The science behind
the software.
Explore the methods behind our tools, the sources that inform them, and the limits that matter when interpreting a result.
A flag starts
an examination.
Flags identify areas or events of interest for further examination. They are not findings of editing, manipulation, or inauthenticity.
An examiner who interprets a flag as evidence of editing must explain the additional examination, supporting observations, alternative explanations considered, and reasoning underlying that conclusion. The flag alone does not support it.
Find the method.
Understand the limits.
Browse by discipline. Open a method to see what it does, its interpretation limits, the development test record and supporting sources.
56 method records across five disciplines
No matching methods. Try a broader term or clear the search.
Image analysis
Examine image structure, repeated content and model-based differences. Results identify observations for examiner review.
Image diagnostics
M01CFA interpolation diagnostic
What the method does
Four Bayer phases; bilinear predictions at interpolated samples; residual RMS and best-versus-second gap; native ROI coordinates.
Interpretation & limitations
Simplified model. Does not implement the full Popescu–Farid EM detector; residuals and gaps are measurements of interest, never confidence or camera identification.
Development test record
Photo fixture log tests all four synthetic phases, odd ROI origin, flat ambiguity, and clipping refusal.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 1 reference
- P1Popescu, A. C., & Farid, H. (2005). Exposing Digital Forgeries in Color Filter Array Interpolated Images. IEEE Transactions on Signal Processing, 53(10), 3948–3959.
Author-hosted manuscript reviewed; CFA method background.
M02Lighting-direction comparison
What the method does
Examiner supplies normals/groups. Inward-offset patch intensity; optional assumed sRGB inverse transfer; Rec.709-weighted intensity; QR least-squares fit of I=LxNx+LyNy+ambient.
Interpretation & limitations
Model-supported simplified variant, not the paper’s full boundary extrapolation/local-light estimator. Shadows, multiple sources, reflectance and tone response matter; angle differences are observations of interest. QR diagonal ratio is not an exact condition number.
Development test record
Photo log reports recovery of a known light vector and angular difference, rejection of rank-deficient inputs.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 1 reference
- P2Johnson, M. K., & Farid, H. (2005). Exposing Digital Forgeries by Detecting Inconsistencies in Lighting. ACM Multimedia and Security Workshop.
Author-hosted paper reviewed; lighting assumptions and model.
M03Focal length / perspective consistency
What the method does
Two orthogonal scene vanishing directions with examiner-supplied principal point: f²=−(v1−c)·(v2−c). Optional sensor-width conversion.
Interpretation & limitations
Uses a geometric constraint from Appendix A, not the paper’s reflection detector. Pinhole, square pixels, zero skew and negligible distortion assumptions; line placement/cropping can dominate differences.
Development test record
Photo log reports known 400-pixel focal-length recovery and parallel-line rejection.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 1 reference
- P3O’Brien, J. F., & Farid, H. (2012). Exposing Photo Manipulation with Inconsistent Reflections. ACM Transactions on Graphics, 31(1). Appendix A.
Author manuscript reviewed; geometric constraints only. Tyree does not implement the paper’s reflection-analysis system.
M04JPEG Error Level Viewer / recompression differences
What the method does
Normalize ROI to 8-bit RGB; re-encode chosen Q, 4:4:4; decode; mean absolute RGB residual per pixel plus cell/RMS summaries and display-only gain.
Interpretation & limitations
Exploratory recompression experiment. Replace P19 as method support: JPEG dimples is a different algorithm. Krawetz is historical non-peer-reviewed context; no exact-variant forensic validation found. Texture, ROI block origin and legitimate processing can explain bright areas.
Development test record
Photo log tests constant/textured fixtures and numeric CSV/hash reproducibility; not sensitivity/specificity.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 3 references
- T1ITU-T T.81 (September 1992). Digital Compression and Coding of Continuous-Tone Still Images—Requirements and Guidelines.
Original standard accessed; JPEG structure/quantization background.
- D3libjpeg-turbo project. src/jcparam.c (rolling upstream source).
IJG quality scaling convention, not universal quality metadata.
- P20Krawetz, N. (2007). A Picture’s Worth… Digital Image Analysis and Forensics. Black Hat USA whitepaper.
Historical ELA reference located from author/conference sources. Conference whitepaper, not peer-reviewed validation; app formula is specified separately.
M05Eight-pixel boundary-phase statistics
What the method does
Mean adjacent luminance-code-value differences grouped by x/y modulo 8 in the selected ROI.
Interpretation & limitations
App-defined descriptive statistic. JPEG block structure supplies context, not a published calibration of this statistic or proof of historical grid/editing.
Development test record
Stored with compression diagnostics; dedicated representative-data validation not established.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 1 reference
- T1ITU-T T.81 (September 1992). Digital Compression and Coding of Continuous-Tone Still Images—Requirements and Guidelines.
Original standard accessed; JPEG structure/quantization background.
M06JPEG quantization inventory
What the method does
Read encoded markers/tables, retain offsets, precision, definitions and hashes; convert stored zigzag order to natural 8×8 order.
Interpretation & limitations
Format-grounded structural inspection. Table IDs alone do not establish color role; missing tables, redefinitions and malformed streams require warnings.
Development test record
Photo log tests natural-order table values against a Pillow-generated fixture.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 1 reference
- T1ITU-T T.81 (September 1992). Digital Compression and Coding of Continuous-Tone Still Images—Requirements and Guidelines.
Original standard accessed; JPEG structure/quantization background.
M07Nominal JPEG quality compatibility
What the method does
Compare each table with IJG luminance/chrominance Q=1–100 tables; exact matches and closest coefficient distance.
Interpretation & limitations
Only compatibility with the IJG convention. Closest distance is not recovered historical quality, an encoder identity or a forgery score; P18’s camera-database method is not implemented.
Development test record
Photo log reports an independent Q91 fixture match.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 3 references
- D3libjpeg-turbo project. src/jcparam.c (rolling upstream source).
IJG quality scaling convention, not universal quality metadata.
- T1ITU-T T.81 (September 1992). Digital Compression and Coding of Continuous-Tone Still Images—Requirements and Guidelines.
Original standard accessed; JPEG structure/quantization background.
- P18Farid, H. (2006). Digital Image Ballistics from JPEG Quantization. Dartmouth Technical Report TR2006-583.
Institutional record/abstract checked. Tyree does not implement its camera/software identification database.
Repeated-content search
M08SIFT extraction and normalization
What the method does
OpenCV SIFT in native-resolution tiles; per-tile budgets; descriptor normalization; grayscale normalization recorded.
Interpretation & limitations
Established component adapted to within-image search. Tiling/budgets and grayscale conversion differ from a general publication pipeline. Low detail and small copies can be missed.
Development test record
Clone harness contains translated, rotated, scaled, JPEG, large-frame and negative fixtures.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 3 references
- P5Lowe, D. G. (2004). Distinctive Image Features from Scale-Invariant Keypoints. International Journal of Computer Vision, 60(2), 91–110.
Author-hosted manuscript reviewed; SIFT and matching foundation.
- P4Christlein, V., Riess, C., Jordan, J., Riess, C., & Angelopoulou, E. (2012). An Evaluation of Popular Copy-Move Forgery Detection Approaches. IEEE TIFS, 7(6), 1841–1854. DOI: 10.1109/TIFS.2012.2218597.
Research record and abstract reviewed; evaluation context, not validation of Tyree’s hybrid.
- N5Lemon, Christopher Douglas (2022). Detecting Cloned Regions in Digital Images Before and After Facebook Recompression. Master’s thesis, University of Colorado Denver.
Repeated-content test corpus and interpretation limits. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
M09Sampled block/patch descriptors
What the method does
Budgeted sampled-patch branch, recorded block size/stride, normalized custom descriptors and low-texture exclusions.
Interpretation & limitations
Block-search family supported; the exact patch descriptor/search rules are app-specific, not a reproduction of every evaluated descriptor. No imported benchmark accuracy claims.
Development test record
Same hybrid fixture records; need branch-isolated evaluation on benign repetition and small/smooth copied areas.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 2 references
- P4Christlein, V., Riess, C., Jordan, J., Riess, C., & Angelopoulou, E. (2012). An Evaluation of Popular Copy-Move Forgery Detection Approaches. IEEE TIFS, 7(6), 1841–1854. DOI: 10.1109/TIFS.2012.2218597.
Research record and abstract reviewed; evaluation context, not validation of Tyree’s hybrid.
- N5Lemon, Christopher Douglas (2022). Detecting Cloned Regions in Digital Images Before and After Facebook Recompression. Master’s thesis, University of Colorado Denver.
Repeated-content test corpus and interpretation limits. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
M10FLANN matching and ratio screening
What the method does
KD-tree matcher, four trees and 96 checks; up to 12 neighbors; distance/separation rules, generalized ratio and near-exact multi-match handling.
Interpretation & limitations
Approximate matching component. App’s generalized ratio/multiple-match logic differs from a single standard Lowe ratio test. The ratio is a matching criterion, not probability.
Development test record
Covered in integrated clone fixtures, not a measured exhaustive-match recall study.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 3 references
- P6Muja, M., & Lowe, D. G. (2009). Fast Approximate Nearest Neighbors with Automatic Algorithm Configuration. VISAPP.
Author-hosted paper reviewed; approximate matching foundation.
- P5Lowe, D. G. (2004). Distinctive Image Features from Scale-Invariant Keypoints. International Journal of Computer Vision, 60(2), 91–110.
Author-hosted manuscript reviewed; SIFT and matching foundation.
- N5Lemon, Christopher Douglas (2022). Detecting Cloned Regions in Digital Images Before and After Facebook Recompression. Master’s thesis, University of Colorado Denver.
Repeated-content test corpus and interpretation limits. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
M11Transform grouping, RANSAC and fit
What the method does
Hough-like scale/angle/translation bins; seeded 192-trial two-point similarity RANSAC; centered least-squares refit; budget and overlap filters.
Interpretation & limitations
Established families with custom selection rules. Similarity only: no reflection/shear/perspective model. Inlier count and residual describe agreement, not editing likelihood.
Development test record
Clone logs report known translation/rotation/scale examples; no representative operational error rates established.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 3 references
- P7Fischler, M. A., & Bolles, R. C. (1981). Random Sample Consensus: A Paradigm for Model Fitting with Applications to Image Analysis and Automated Cartography. Communications of the ACM, 24(6), 381–395.
Originating institution citation checked; generic RANSAC, not app-specific acceptance rules.
- P4Christlein, V., Riess, C., Jordan, J., Riess, C., & Angelopoulou, E. (2012). An Evaluation of Popular Copy-Move Forgery Detection Approaches. IEEE TIFS, 7(6), 1841–1854. DOI: 10.1109/TIFS.2012.2218597.
Research record and abstract reviewed; evaluation context, not validation of Tyree’s hybrid.
- N5Lemon, Christopher Douglas (2022). Detecting Cloned Regions in Digital Images Before and After Facebook Recompression. Master’s thesis, University of Colorado Denver.
Repeated-content test corpus and interpretation limits. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
M12Local correlation and support overlays
What the method does
Bilinear sampled 13×13 neighborhoods; normalized correlation; support hulls/grouped highlights.
Interpretation & limitations
Correlation checks visual agreement. Hulls are support extents, not copied boundaries; natural repetition can pass. Exact acceptance/display rules are app-specific.
Development test record
Existing exports and coordinate/highlight harnesses; not exact segmentation validation.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 2 references
- P4Christlein, V., Riess, C., Jordan, J., Riess, C., & Angelopoulou, E. (2012). An Evaluation of Popular Copy-Move Forgery Detection Approaches. IEEE TIFS, 7(6), 1841–1854. DOI: 10.1109/TIFS.2012.2218597.
Research record and abstract reviewed; evaluation context, not validation of Tyree’s hybrid.
- N5Lemon, Christopher Douglas (2022). Detecting Cloned Regions in Digital Images Before and After Facebook Recompression. Master’s thesis, University of Colorado Denver.
Repeated-content test corpus and interpretation limits. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
Image & video processing
Understand how enhancement, restoration and geometric transforms change the displayed or exported image.
Enhancement & restoration
M13Brightness, contrast and gamma
What the method does
Video uses clipped 8-bit LUT with half-up rounding; still images use range-normalized float64 arithmetic.
Interpretation & limitations
Conventional point transforms. Parameters alter code values; they do not recover calibrated scene radiance or true color. Image and video precision paths are different.
Development test record
Existing LUT/pipeline parity fixtures; broad source/color coverage remains separate.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 2 references
- B1Smith, S. W. (1997). The Scientist and Engineer’s Guide to Digital Signal Processing, chapter 24, Linear Image Processing.
Author-hosted textbook chapter reviewed; convolution, smoothing, sharpening and restoration families.
- B3Gonzalez, R. C., & Woods, R. E. (2018). Digital Image Processing, fourth edition. Pearson.
Author textbook metadata/scope checked. Full book/page-level verification not performed; use as broad background only.
M14Levels, curves and manual color gains
What the method does
Video LUT/FFmpeg curves, channel gains and saturation; image levels/range arithmetic and RGB gains/saturation.
Interpretation & limitations
Conventional mapping, not automatic illuminant estimation. Gains/clipping/colorspace conversions can alter details; video curves are not implied in the image UI.
Development test record
Release/pipeline harnesses; not a colorimetric validation.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 2 references
- B3Gonzalez, R. C., & Woods, R. E. (2018). Digital Image Processing, fourth edition. Pearson.
Author textbook metadata/scope checked. Full book/page-level verification not performed; use as broad background only.
- D1FFmpeg project. Filters Documentation (rolling reference, checked November 10, 2026).
Implementation documentation. Current documentation must be reconciled with shipped build.
M15Color-space conversion and orientation
What the method does
Explicit 16-bit sRGB conversion via ICC/libvips where requested; source orientation and alpha handling; metadata recorded.
Interpretation & limitations
Standards-backed workflow, not display calibration. Native samples and interpreted display must remain distinguished; inspect actual profile/library behavior.
Development test record
Existing image precision/orientation fixtures, not full ICC conformance testing.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 2 references
- T6International Color Consortium. ICC.1 v4 specification landing page; existing guide cites ICC.1:2022.
Official source located; exact library conformance and display calibration not established.
- N4Kost, Susan Clarisa (2025). A Quantitative Study of Frame Extraction Methods, Tools, and Their Impact on Extracted Frame Quality from HEVC Videos. Master’s thesis, University of Colorado Denver.
Extraction/conversion test design and representation limits. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
M16Binomial/Gaussian/box smoothing primitives
What the method does
Video has integer/binomial and quantized Gaussian/box operators with explicit borders and rounding; still-image Gaussian is separable float64.
Interpretation & limitations
Established convolution families. Quantization, support, boundary and intermediate rounding are implementation definitions requiring oracle checks.
Development test record
Restoration/unsharp notes document CPU/GPU/export parity on controlled fixtures.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 1 reference
- B1Smith, S. W. (1997). The Scientist and Engineer’s Guide to Digital Signal Processing, chapter 24, Linear Image Processing.
Author-hosted textbook chapter reviewed; convolution, smoothing, sharpening and restoration families.
M17Selective spatial noise reduction
What the method does
Blend toward blur only when channel residual is below threshold; integer binomial video versus float64 Gaussian image path.
Interpretation & limitations
Custom thresholded smoothing, not a validated classifier of noise versus detail or a claim of bilateral/NLM processing. Threshold/strength can suppress real detail.
Development test record
Synthetic noise and parity fixtures documented; operational noise/detail tradeoffs not established.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 1 reference
- B1Smith, S. W. (1997). The Scientist and Engineer’s Guide to Digital Signal Processing, chapter 24, Linear Image Processing.
Author-hosted textbook chapter reviewed; convolution, smoothing, sharpening and restoration families.
M18Unsharp Mask
What the method does
Boost input-minus-blur with amount/threshold; video offers binomial/Gaussian/box choices; images use float64 Gaussian.
Interpretation & limitations
Conventional high-boost family with specific threshold/rounding rules. Halos, clipping and noise can create misleading apparent structure.
Development test record
Documented direct-convolution oracle, small-image/border, CPU/GPU/export and alpha tests.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 1 reference
- B1Smith, S. W. (1997). The Scientist and Engineer’s Guide to Digital Signal Processing, chapter 24, Linear Image Processing.
Author-hosted textbook chapter reviewed; convolution, smoothing, sharpening and restoration families.
M19Local contrast
What the method does
Wide Gaussian high boost; video graph and still-image float64 implementation have different parameters/precision.
Interpretation & limitations
Neighborhood contrast enhancement, not CLAHE. It can amplify texture/noise and introduce halos.
Development test record
Release/pipeline fixtures; no perceptual or forensic superiority established.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 1 reference
- B1Smith, S. W. (1997). The Scientist and Engineer’s Guide to Digital Signal Processing, chapter 24, Linear Image Processing.
Author-hosted textbook chapter reviewed; convolution, smoothing, sharpening and restoration families.
M20Histogram equalization
What the method does
Video FFmpeg histeq; still images use per-band 65,536-bin CDF mapping blended by strength.
Interpretation & limitations
Conventional histogram-family transforms, not identical across workspaces. Per-band mapping can alter color and evidence appearance.
Development test record
Implementation fixtures exist; no claim of calibrated contrast recovery.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 2 references
- B3Gonzalez, R. C., & Woods, R. E. (2018). Digital Image Processing, fourth edition. Pearson.
Author textbook metadata/scope checked. Full book/page-level verification not performed; use as broad background only.
- D1FFmpeg project. Filters Documentation (rolling reference, checked November 10, 2026).
Implementation documentation. Current documentation must be reconciled with shipped build.
M21Van Cittert deblur
What the method does
Residual update x←x+s(y−Hx); video clips/rounds each iteration; image uses float64; selected PSF recorded.
Interpretation & limitations
Specified variants, not general convergence guarantees. Legacy box/line PSFs differ from later directional Landweber behavior; retain exact method ID.
Development test record
Controlled model-matched and export-parity checks in method notes.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 2 references
- P10Hill, N. R., & Ioup, G. E. (1976). Convergence of the van Cittert Iterative Method of Deconvolution. JOSA, 66(5), 487–489. DOI: 10.1364/JOSA.66.000487.
Publisher record/abstract checked. Convergence is conditional; this is not the original historical invention paper.
- P12Sage, D., et al. (2017). DeconvolutionLab2: An Open-Source Software for Deconvolution Microscopy. Methods, 115, 28–41.
Author-hosted paper reviewed; Van Cittert, Landweber, regularized inversion and Richardson–Lucy background.
M22Directional-motion Landweber-style deblur
What the method does
Centered directional line PSF; residual update using H(y)−H(H(x)); integer quantization for video versus double image arithmetic.
Interpretation & limitations
With edge replication, the finite-image operator is not automatically its exact adjoint. Quantization/boundaries prevent transferring theoretical convergence claims unchanged.
Development test record
Documented directional/transpose/opposite-angle/parity fixtures.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 2 references
- P11Landweber, L. (1951). An Iteration Formula for Fredholm Integral Equations of the First Kind. American Journal of Mathematics, 73(3), 615–624. DOI: 10.2307/2372313.
Original bibliographic identity corroborated by the EPFL paper; publisher full text inaccessible in this review.
- P12Sage, D., et al. (2017). DeconvolutionLab2: An Open-Source Software for Deconvolution Microscopy. Methods, 115, 28–41.
Author-hosted paper reviewed; Van Cittert, Landweber, regularized inversion and Richardson–Lucy background.
M23Richardson–Lucy deblur
What the method does
Positive float32 multiplicative iteration; epsilon 10⁻⁶; relaxation exponent; symmetric periodic PSF convolution after replicated-edge padding, then RGB24 output.
Interpretation & limitations
Relaxed app variant, not unmodified original algorithm. Poisson measurement assumptions are not established for compressed RGB; PSF mismatch and iterations affect artifacts. Not present in the reviewed still-image pipeline.
Development test record
New-method notes report raw-preview/cache/lossless-export agreement and controlled point/flat-field fixtures.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 3 references
- P8Richardson, W. H. (1972). Bayesian-Based Iterative Method of Image Restoration. JOSA, 62(1), 55–59. DOI: 10.1364/JOSA.62.000055.
Publisher record and abstract checked; full-text equivalence not established.
- P9Lucy, L. B. (1974). An Iterative Technique for the Rectification of Observed Distributions. Astronomical Journal, 79, 745–754.
Original paper hosted by university reviewed.
- P12Sage, D., et al. (2017). DeconvolutionLab2: An Open-Source Software for Deconvolution Microscopy. Methods, 115, 28–41.
Author-hosted paper reviewed; Van Cittert, Landweber, regularized inversion and Richardson–Lucy background.
M24Wiener-style regularized inverse
What the method does
FFmpeg frequency-domain deconvolve, centered quantized RGB16 PSF, constant K and source/restored strength blend with edge padding/crop.
Interpretation & limitations
Regularized inverse family; K is selected, not an estimated signal/noise spectrum. Do not claim a measured optimal Wiener solution for arbitrary evidence. Not present in the reviewed still-image pipeline.
Development test record
New-method notes report pixel-path agreement and parameter-sensitive output; no broad restoration validation.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 3 references
- P13Wiener, N. (1949). Extrapolation, Interpolation, and Smoothing of Stationary Time Series: With Engineering Applications. MIT Press.
Historical foundation; publisher search record and EPFL bibliography checked. Book full text not obtained.
- P12Sage, D., et al. (2017). DeconvolutionLab2: An Open-Source Software for Deconvolution Microscopy. Methods, 115, 28–41.
Author-hosted paper reviewed; Van Cittert, Landweber, regularized inversion and Richardson–Lucy background.
- D2FFmpeg project. vf_convolve.c (rolling upstream source).
Deconvolution implementation reference; not independently validated Tyree behavior.
Temporal processing
M25Arithmetic frame averaging
What the method does
Equal-weight decoded gamma-encoded RGB8 mean, unsigned sums and half-up rounding, inclusive frame range.
Interpretation & limitations
Combines frames, not newly measured detail. Motion can ghost; equal frame weighting differs from duration weighting and linear-light averaging.
Development test record
Independent full-decode/rounding/source-preservation checks documented.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 1 reference
- B4Smith, S. W. (1997). The Scientist and Engineer’s Guide to Digital Signal Processing, chapter 25, Special Imaging Techniques.
Author-hosted chapter accessed; image SNR and averaging context.
M26Integer-translation alignment before averaging
What the method does
Exhaustive limited shift search versus first selected frame; sampled grayscale mean absolute difference, shortest-shift tie break; average valid contributors at borders.
Interpretation & limitations
App-defined registration wrapper. No subpixel/rotation/object motion or calibrated confidence; weak texture and moving objects can cause wrong alignment. No exact published-validation match found.
Development test record
Known-translation fixture documented; representative motion/occlusion testing remains required.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 1 reference
- B4Smith, S. W. (1997). The Scientist and Engineer’s Guide to Digital Signal Processing, chapter 25, Special Imaging Techniques.
Author-hosted chapter accessed; image SNR and averaging context.
M27BWDIF deinterlacing
What the method does
FFmpeg bwdif send_frame, selected field order, forced deinterlacing.
Interpretation & limitations
Backend-defined processing, not recovery of every original temporal sample. Field-order errors and interpolation can change structure; pin backend source.
Development test record
Release/on-demand fixtures compare selected frames and exported results; real interlaced corpus not established here.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 1 reference
- D1FFmpeg project. Filters Documentation (rolling reference, checked November 10, 2026).
Implementation documentation. Current documentation must be reconciled with shipped build.
M28HQDN3D temporal denoising
What the method does
FFmpeg HQDN3D with near-zero spatial strengths and selected temporal strength, through YUV444 conversions.
Interpretation & limitations
Exact implementation source is stronger support than assigning an unrelated denoising paper. Motion/detail effects and conversions require testing; peer-reviewed validation of this Tyree configuration not located.
Development test record
Temporal/export/on-demand parity fixtures documented.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 1 reference
- D1FFmpeg project. Filters Documentation (rolling reference, checked November 10, 2026).
Implementation documentation. Current documentation must be reconciled with shipped build.
M29Deshake stabilization
What the method does
FFmpeg global frame-to-frame stabilization, selected search range and blank edges.
Interpretation & limitations
Backend method, not object tracking, rolling-shutter correction or proof of physical motion. Exact build and initialization matter.
Development test record
Stateful preview/export fixtures documented; no operational accuracy characterization located.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 1 reference
- D1FFmpeg project. Filters Documentation (rolling reference, checked November 10, 2026).
Implementation documentation. Current documentation must be reconciled with shipped build.
Geometry & analytical views
M30Crop, rotation, flips and resize
What the method does
Discrete geometric transforms plus selected video interpolation; still-image alpha-aware bilinear resize.
Interpretation & limitations
Pixel mapping/resampling, not new measured detail. Image interpolation/coordinates and video geometry must be documented separately.
Development test record
Existing geometry/pipeline fixtures; no calibrated dimensional-measurement claim.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 4 references
- G5SWGDE. Fundamentals of Resizing Imagery and Considerations for Legal Proceedings, 22-V-001, version 1.1 (December 16, 2025), final; cited edition.
Official document accessed; resampling context.
- B3Gonzalez, R. C., & Woods, R. E. (2018). Digital Image Processing, fourth edition. Pearson.
Author textbook metadata/scope checked. Full book/page-level verification not performed; use as broad background only.
- N4Kost, Susan Clarisa (2025). A Quantitative Study of Frame Extraction Methods, Tools, and Their Impact on Extracted Frame Quality from HEVC Videos. Master’s thesis, University of Colorado Denver.
Extraction/conversion test design and representation limits. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
- N6Latham, Jason (2024). Evaluation of Interpolation Methods and Their Impact on Resizing Images for Use in Digital and Multimedia Evidence (DME) Forensics—Is the Nearest Neighbor Algorithm the Best Solution in All Situations? Master’s thesis, University of Colorado Denver.
Interpolation documentation and targeted test design. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
M31Perspective rectification
What the method does
Examiner’s convex quadrilateral mapped to a rectangle with cubic interpolation.
Interpretation & limitations
Image warp, not calibrated photogrammetry, recovered dimensions or the separate focal-length analysis.
Development test record
Degenerate/crossing-geometry and export fixtures documented.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 2 references
- B3Gonzalez, R. C., & Woods, R. E. (2018). Digital Image Processing, fourth edition. Pearson.
Author textbook metadata/scope checked. Full book/page-level verification not performed; use as broad background only.
- N6Latham, Jason (2024). Evaluation of Interpolation Methods and Their Impact on Resizing Images for Use in Digital and Multimedia Evidence (DME) Forensics—Is the Nearest Neighbor Algorithm the Best Solution in All Situations? Master’s thesis, University of Colorado Denver.
Interpolation documentation and targeted test design. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
M32Channel/grayscale views
What the method does
Selected band replication or weighted RGB view; image alpha view separately supported.
Interpretation & limitations
Descriptive view. Fixed weights on code values do not establish measured luminance or calibrated radiance.
Development test record
Existing channel/pipeline tests.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 1 reference
- B3Gonzalez, R. C., & Woods, R. E. (2018). Digital Image Processing, fourth edition. Pearson.
Author textbook metadata/scope checked. Full book/page-level verification not performed; use as broad background only.
M33Canny-style video edges
What the method does
FFmpeg edgedetect wires mode with explicit low/high thresholds.
Interpretation & limitations
Edge response, not confirmed object boundaries; exact backend differs from every detail of the original multiscale paper.
Development test record
Release fixtures exercise edge path, not object-boundary truth.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 2 references
- P17Canny, J. (1986). A Computational Approach to Edge Detection. IEEE TPAMI, 8(6), 679–698.
Original paper reviewed; family context for video edge view, not the still-image Sobel operation.
- D1FFmpeg project. Filters Documentation (rolling reference, checked November 10, 2026).
Implementation documentation. Current documentation must be reconciled with shipped build.
M34Sobel still-image edges
What the method does
Custom float64 3×3 horizontal/vertical Sobel magnitude divided by four with replicated boundaries.
Interpretation & limitations
Conventional derivative operator, separate from Canny-style video path. Noise/contrast affects results; not an object identification test.
Development test record
Image implementation fixtures, not calibrated boundary detection.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 1 reference
- D4OpenCV 4.13.0. Sobel Derivatives.
Operator reference; Tyree still-image implementation is custom.
Audio analysis & processing
Keep original-source observations, signal processing and visual displays distinct. Each serves a different purpose.
Original-source observations
M43Amplitude-discontinuity events of interest
What the method does
Absolute first difference divided by preceding exponentially weighted difference RMS; jump/sensitivity/separation thresholds; first 20 ms excluded.
Interpretation & limitations
First-derivative examination has guidance support. App’s adaptive denominator/threshold rules lack an established operational calibration. Output must be events of interest, never confirmed butt splices, confidence or editing probability.
Development test record
Historical log reports detection of a seeded jump and sample/time relation; natural-transient and smooth-edit miss testing remains important.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 3 references
- G2SWGDE. Best Practices for Digital Audio Authentication, 15-A-001, version 1.3 (September 20, 2018), final, sections 4.5.1–4.5.4.
Waveform, spectral and first-derivative examination guidance checked. Not a calibration of Tyree thresholds.
- N1Aeschliman, Austin Jon (2020). Butt-Splice Detection on Audio Recordings with Lossy Compression. Master’s thesis, University of Colorado Denver.
Method comparison, experimental design and limitations; not validation of Tyree. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
- N2Pruett, Shawn Michael (2019). An Exploratory Study of Manual Pause-Record Events in Digital Audio Recordings. Master’s thesis, University of Colorado Denver.
Alternative explanations and validation-corpus design. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
M44Near-full-scale runs of interest
What the method does
Runs of at least three decoded samples with |sample|≥0.9999; explicit result caps.
Interpretation & limitations
App-defined threshold screen, not proof of clipping; analog or lower-level clipping may be missed.
Development test record
Current source documents behavior; representative clipping-classification validation not found.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 3 references
- G2SWGDE. Best Practices for Digital Audio Authentication, 15-A-001, version 1.3 (September 20, 2018), final, sections 4.5.1–4.5.4.
Waveform, spectral and first-derivative examination guidance checked. Not a calibration of Tyree thresholds.
- N1Aeschliman, Austin Jon (2020). Butt-Splice Detection on Audio Recordings with Lossy Compression. Master’s thesis, University of Colorado Denver.
Method comparison, experimental design and limitations; not validation of Tyree. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
- N2Pruett, Shawn Michael (2019). An Exploratory Study of Manual Pause-Record Events in Digital Audio Recordings. Master’s thesis, University of Colorado Denver.
Alternative explanations and validation-corpus design. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
Signal processing
M35High/low-pass and parametric EQ
What the method does
FFmpeg highpass/lowpass and three configurable equalizer bands; Nyquist checks.
Interpretation & limitations
Conventional frequency filtering; response, phase and attenuation can remove wanted information. No speaker identification or authenticity determination.
Development test record
Existing audio log tests attenuation of 60 Hz and retention of 1 kHz.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 3 references
- B2Smith, S. W. (1997). The Scientist and Engineer’s Guide to Digital Signal Processing, chapter 19, Recursive Filters.
Author-hosted textbook chapter reviewed; recursive-filter foundation.
- D1FFmpeg project. Filters Documentation (rolling reference, checked November 10, 2026).
Implementation documentation. Current documentation must be reconciled with shipped build.
- N7Zjalic, James (2017). A Proposed Framework for Forensic Audio Enhancement. Master’s thesis, University of Colorado Denver.
Workflow framework and outcome interpretation; no universal optimum. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
M36Hum/harmonic notches
What the method does
Configurable equalizer cuts at selected fundamental/harmonics below Nyquist.
Interpretation & limitations
Filtering of selected frequencies, not ENF-based recording authentication or mains history matching.
Development test record
Synthetic hum/tone fixture documented.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 3 references
- B2Smith, S. W. (1997). The Scientist and Engineer’s Guide to Digital Signal Processing, chapter 19, Recursive Filters.
Author-hosted textbook chapter reviewed; recursive-filter foundation.
- D1FFmpeg project. Filters Documentation (rolling reference, checked November 10, 2026).
Implementation documentation. Current documentation must be reconciled with shipped build.
- N7Zjalic, James (2017). A Proposed Framework for Forensic Audio Enhancement. Master’s thesis, University of Colorado Denver.
Workflow framework and outcome interpretation; no universal optimum. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
M37Broadband denoising and learned noise profile
What the method does
FFmpeg afftdn fixed/adaptive/custom profile; 15-band per-channel profiles averaged, context-bound to source/track/upstream recipe.
Interpretation & limitations
Backend-defined spectral processing. Learned profile means measured noise-only segment, not AI training. No exact peer-reviewed basis for all backend/app profile rules established; wanted signal may be attenuated.
Development test record
Noise-only energy/tone retention/profile invalidation fixtures in historical log.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 3 references
- D1FFmpeg project. Filters Documentation (rolling reference, checked November 10, 2026).
Implementation documentation. Current documentation must be reconciled with shipped build.
- G3SWGDE. Best Practices for Enhancement of Digital Audio, 20-A-001, version 2.0 (August 21, 2025), final; cited edition.
Official document accessed; workflow guidance, not exact-filter validation.
- N7Zjalic, James (2017). A Proposed Framework for Forensic Audio Enhancement. Master’s thesis, University of Colorado Denver.
Workflow framework and outcome interpretation; no universal optimum. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
M38Declipping
What the method does
FFmpeg adeclip with explicit threshold/window/autoregression-order parameters.
Interpretation & limitations
AR reconstruction family; reconstructed samples are estimates, not recovered known originals. No exact equivalence to Janssen et al. established.
Development test record
Controlled clipped-sine fixture reports improved MSE only for that experiment.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 3 references
- P15Janssen, A. J. E. M., Veldhuis, R. N. J., & Vries, L. B. (1986). Adaptive Interpolation of Discrete-Time Signals That Can Be Modeled as Autoregressive Processes. IEEE TASSP, 34(2), 317–330.
Author institution record/abstract reviewed; AR interpolation family, not proof FFmpeg implements this paper exactly.
- D1FFmpeg project. Filters Documentation (rolling reference, checked November 10, 2026).
Implementation documentation. Current documentation must be reconciled with shipped build.
- N7Zjalic, James (2017). A Proposed Framework for Forensic Audio Enhancement. Master’s thesis, University of Colorado Denver.
Workflow framework and outcome interpretation; no universal optimum. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
M39Click/impulse reduction
What the method does
FFmpeg adeclick with selected threshold/window/burst fusion.
Interpretation & limitations
AR interpolation family with backend detection/reconstruction. Natural transients may be changed; this is not the original-source discontinuity scan.
Development test record
Synthetic impulse attenuation documented; representative speech/transient testing still needed.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 3 references
- P15Janssen, A. J. E. M., Veldhuis, R. N. J., & Vries, L. B. (1986). Adaptive Interpolation of Discrete-Time Signals That Can Be Modeled as Autoregressive Processes. IEEE TASSP, 34(2), 317–330.
Author institution record/abstract reviewed; AR interpolation family, not proof FFmpeg implements this paper exactly.
- D1FFmpeg project. Filters Documentation (rolling reference, checked November 10, 2026).
Implementation documentation. Current documentation must be reconciled with shipped build.
- N7Zjalic, James (2017). A Proposed Framework for Forensic Audio Enhancement. Master’s thesis, University of Colorado Denver.
Workflow framework and outcome interpretation; no universal optimum. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
M40Fixed gain and peak normalization
What the method does
Double-precision volume gain; peak measurement via astats feeds explicit dB adjustment.
Interpretation & limitations
Deterministic level manipulation. Peak normalization is not loudness normalization, calibrated SPL or proof of unchanged signal.
Development test record
Fixed dB/channel isolation and measured target-peak fixtures documented.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 2 references
- D1FFmpeg project. Filters Documentation (rolling reference, checked November 10, 2026).
Implementation documentation. Current documentation must be reconciled with shipped build.
- N7Zjalic, James (2017). A Proposed Framework for Forensic Audio Enhancement. Master’s thesis, University of Colorado Denver.
Workflow framework and outcome interpretation; no universal optimum. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
M41Loudness normalization
What the method does
FFmpeg loudnorm; adjustable target, TP −1, LRA 11 and dual_mono false.
Interpretation & limitations
Standards-related level processing; no product conformance certification. Settings may differ from a broadcast delivery specification.
Development test record
Historical log exercises this mode; formal reference-vector compliance not established.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 4 references
- T3ITU-R BS.1770-5 (November 2023). Algorithms to Measure Audio Programme Loudness and True-Peak Audio Level.
Official current/superseded-version listing checked. App conformance not tested here.
- T4EBU R 128, version 5.0 (November 2023). Loudness Normalisation and Permitted Maximum Level of Audio Signals.
Official recommendation checked; app exposes adjustable loudness target.
- D1FFmpeg project. Filters Documentation (rolling reference, checked November 10, 2026).
Implementation documentation. Current documentation must be reconciled with shipped build.
- N7Zjalic, James (2017). A Proposed Framework for Forensic Audio Enhancement. Master’s thesis, University of Colorado Denver.
Workflow framework and outcome interpretation; no universal optimum. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
M42Dynamic compression
What the method does
FFmpeg acompressor threshold/ratio/attack/release, makeup 1, maximum channel linking.
Interpretation & limitations
Dynamics family, not pristine recovery of original speech dynamics; exact transfer/ballistics depend on backend.
Development test record
Synthetic above-threshold reduction documented.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 3 references
- P16Giannoulis, D., Massberg, M., & Reiss, J. D. (2012). Digital Dynamic Range Compressor Design—A Tutorial and Analysis. JAES, 60(6), 399–408.
Publisher citation checked; compressor design background.
- D1FFmpeg project. Filters Documentation (rolling reference, checked November 10, 2026).
Implementation documentation. Current documentation must be reconciled with shipped build.
- N7Zjalic, James (2017). A Proposed Framework for Forensic Audio Enhancement. Master’s thesis, University of Colorado Denver.
Workflow framework and outcome interpretation; no universal optimum. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
Waveforms & spectrograms
M45Waveform envelopes
What the method does
Per-channel native-rate float32 decode to per-column min/max in workspace; legacy overlay uses FFmpeg showwavespic.
Interpretation & limitations
Time/amplitude display; overview bins cannot substitute for individual-sample examination. Probed versus decoded durations and renderer differences need testing.
Development test record
Waveform/ruler/export harnesses exist; display is not calibrated acoustic amplitude.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 2 references
- G2SWGDE. Best Practices for Digital Audio Authentication, 15-A-001, version 1.3 (September 20, 2018), final, sections 4.5.1–4.5.4.
Waveform, spectral and first-derivative examination guidance checked. Not a calibration of Tyree thresholds.
- N2Pruett, Shawn Michael (2019). An Exploratory Study of Manual Pause-Record Events in Digital Audio Recordings. Master’s thesis, University of Colorado Denver.
Alternative explanations and validation-corpus design. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
M46Spectrogram and coordinate mapping
What the method does
FFmpeg showspectrumpic plus selected window, gain, dynamic range, interval/frequency mapping and axes; legacy and shared renderer paths differ.
Interpretation & limitations
Windowed spectral view. Relative FFT display is not calibrated SPL or a direct amplitude measurement; resolution and mappings depend on render settings.
Development test record
Spectrum/axis/audio-video export harnesses exist; exact FFT length/overlap per backend remains to be pinned.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 5 references
- P14Harris, F. J. (1978). On the Use of Windows for Harmonic Analysis with the Discrete Fourier Transform. Proceedings of the IEEE, 66(1), 51–83.
Original paper reviewed; windowing and spectral interpretation.
- G2SWGDE. Best Practices for Digital Audio Authentication, 15-A-001, version 1.3 (September 20, 2018), final, sections 4.5.1–4.5.4.
Waveform, spectral and first-derivative examination guidance checked. Not a calibration of Tyree thresholds.
- D1FFmpeg project. Filters Documentation (rolling reference, checked November 10, 2026).
Implementation documentation. Current documentation must be reconciled with shipped build.
- N1Aeschliman, Austin Jon (2020). Butt-Splice Detection on Audio Recordings with Lossy Compression. Master’s thesis, University of Colorado Denver.
Method comparison, experimental design and limitations; not validation of Tyree. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
- N2Pruett, Shawn Michael (2019). An Exploratory Study of Manual Pause-Record Events in Digital Audio Recordings. Master’s thesis, University of Colorado Denver.
Alternative explanations and validation-corpus design. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
Integrity & inspection
Inspect file structure and provenance, and verify byte-level integrity. These checks address different questions from scene authenticity.
File inspection & provenance
M47SHA-256 integrity and audit chaining
What the method does
Source/output SHA-256; serialized audit-entry previous-hash chaining and project consistency hash.
Interpretation & limitations
Standard hash foundation. A locally recomputable unkeyed chain is not an independent trusted timestamp, signature, or proof of scene authenticity.
Development test record
Known-hash/source-preservation/project tamper fixtures exist; certification is not established.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 3 references
- T2NIST FIPS 180-4 (August 2015). Secure Hash Standard. DOI: 10.6028/NIST.FIPS.180-4.
Standard accessed; SHA family. No NIST certification of Tyree asserted.
- N3Smyly, John Woodward (2023). Investigating the Stream Hashing Function in FFmpeg. Master’s thesis, University of Colorado Denver.
Reproducibility and hash-domain controls; no claim of exact Tyree reproduction. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
- D8FFmpeg project. Formats Documentation, hash and streamhash muxers (rolling documentation, reviewed November 10, 2026).
Hash domain and default decoded-sample representation; inspect the pinned executable rather than assuming documentation equals installed behavior.
M48Legacy and comparison hashes
What the method does
Tyree Lite/Grab and Hash provide MD5, SHA-1, SHA-256, SHA-384 and SHA-512 comparison through platform cryptography; see variant sources in inventory.
Interpretation & limitations
SHA-256 remains integrity baseline. MD5/SHA-1 are compatibility comparisons, not collision-resistant assurances. No blanket FIPS-certified product claim.
Development test record
Variant checks/documents describe expected-hash comparison; source filenames differ by project.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 1 reference
- T2NIST FIPS 180-4 (August 2015). Secure Hash Standard. DOI: 10.6028/NIST.FIPS.180-4.
Standard accessed; SHA family. No NIST certification of Tyree asserted.
M49MediaInfo, FFprobe, EXIF and JPEG inspection
What the method does
Library/format-based inspection, typed metadata and structural inventories; native adapters live in UI projects.
Interpretation & limitations
Parsing and measurement. Tags can be wrong or edited; displayed values do not establish actual device, recording date or authenticity. Parser subset and errors matter.
Development test record
EXIF/inspection/variant fixture records exist; full format conformance not established.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 4 references
- D5FFmpeg project. FFprobe Documentation.
Stream/frame metadata interpretation interface.
- D6MediaArea. MediaInfo official documentation/product description.
Parser implementation reference; metadata is not verified ground truth.
- T7CIPA. Exif standards catalog.
Official catalog checked. Parser supports a subset, not claimed full support for a particular latest Exif edition.
- T1ITU-T T.81 (September 1992). Digital Compression and Coding of Continuous-Tone Still Images—Requirements and Guidelines.
Original standard accessed; JPEG structure/quantization background.
M50Hex and byte/string search
What the method does
Read-only paged bytes and bounded keyword/hex matching.
Interpretation & limitations
Software correctness task rather than an authenticity algorithm; offsets/encodings/buffer boundaries require known-answer tests. No special scientific accuracy claim.
Development test record
Hex-search harness exists.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 1 reference
- G1SWGDE. Minimum Requirements for Testing Tools Used in Digital and Multimedia Forensics, 18-Q-001, version 2.1 (March 7, 2024), final.
Testing-category guidance accessed; no endorsement/certification.
M51AI-related metadata and Content Credentials
What the method does
Field-specific generator-name rules plus c2patool active signature/file-binding/trust interpretation; local trust snapshot, no pixel/audio classification.
Interpretation & limitations
Provenance screen only. Missing clues do not prove non-AI; signed assertions are not independent truth. Historical/ingredient data and untrusted signatures require separate labels.
Development test record
Existing rule/credential behavior visible in source; exact shipped-validator trust/revocation coverage needs version-specific tests.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 2 references
- T5C2PA Technical Specification, version 2.2 (cited edition).
Cited edition accessed; do not call it latest or infer complete validator conformance.
- D7Content Authenticity Initiative. c2patool release v0.28.6.
Release reference matched to local dependency note; executable not independently reverified here.
Acquisition & exhibits
Follow the path from source acquisition to extracted frames, labels, demonstrative exhibits and exported media.
Source handling & presentation
M52Online media acquisition
What the method does
yt-dlp child process; acquisition records and local file hashes.
Interpretation & limitations
Workflow/tool documentation, not proof a remote recording is original or complete. Site/extractor/version/network behavior affects what is obtained.
Development test record
Variant acquisition records exist; no universal website acquisition validation.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 1 reference
- G1SWGDE. Minimum Requirements for Testing Tools Used in Digital and Multimedia Forensics, 18-Q-001, version 2.1 (March 7, 2024), final.
Testing-category guidance accessed; no endorsement/certification.
M53Frame extraction and timing
What the method does
Original-suite decoded-frame PNG export. Lite/Grab and Hash have independent indexed extraction using original frame order/PTS/time base, output hashes and completion checks.
Interpretation & limitations
Decoded frame index/PTS is not authenticated wall-clock recording time. RGB24 conversion changes representation and may lose HDR/high-bit-depth information; viewing proxies must not replace evidence.
Development test record
Frame/first-frame and Lite/Grab and Hash checks exist; exact packaged builds require separate validation.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 5 references
- D5FFmpeg project. FFprobe Documentation.
Stream/frame metadata interpretation interface.
- G4SWGDE. Best Practices for Digital Forensic Video Analysis, 18-V-001, version 2.0 (December 10, 2025), final; cited edition.
Official document accessed; workflow context.
- N1Aeschliman, Austin Jon (2020). Butt-Splice Detection on Audio Recordings with Lossy Compression. Master’s thesis, University of Colorado Denver.
Method comparison, experimental design and limitations; not validation of Tyree. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
- N4Kost, Susan Clarisa (2025). A Quantitative Study of Frame Extraction Methods, Tools, and Their Impact on Extracted Frame Quality from HEVC Videos. Master’s thesis, University of Colorado Denver.
Extraction/conversion test design and representation limits. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
- N6Latham, Jason (2024). Evaluation of Interpolation Methods and Their Impact on Resizing Images for Use in Digital and Multimedia Evidence (DME) Forensics—Is the Nearest Neighbor Algorithm the Best Solution in All Situations? Master’s thesis, University of Colorado Denver.
Interpolation documentation and targeted test design. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
M54Visible timestamps and frame counters
What the method does
User/configuration-derived time and decoded index overlays; annotations use recorded frame ranges.
Interpretation & limitations
Presentation labels, not authenticated acquisition timestamps. User assumptions, source PTS and export/system dates must remain distinct.
Development test record
Timestamp/frame-range fixtures exist.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 1 reference
- G4SWGDE. Best Practices for Digital Forensic Video Analysis, 18-V-001, version 2.0 (December 10, 2025), final; cited edition.
Official document accessed; workflow context.
M55Annotations, redaction and exhibits
What the method does
Boxes/text/arrows and opaque marks; separate Exhibit Builder model/rendering; timed intervals and source rehashing on reopen.
Interpretation & limitations
Demonstrative graphics, not analytical findings. Comparison/export can expose unredacted originals; marks do not sanitize unrelated streams or metadata. Exhibit project is independent and currently lacks exhibit export.
Development test record
Exhibit model/rendering checks and source README reviewed; inherited hidden analysis code is not an exposed feature.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 1 reference
- G4SWGDE. Best Practices for Digital Forensic Video Analysis, 18-V-001, version 2.0 (December 10, 2025), final; cited edition.
Official document accessed; workflow context.
M56Lossless masters, presentation exports and verification
What the method does
Selected encoders, package source/output hashes and, in relevant converted-audio paths, canonical decoded PCM hash comparisons; lossy viewing/presentation paths separately recorded.
Interpretation & limitations
Lossless encoding preserves the supplied processed representation, not pre-processing information. File hashes and decoded-sample hashes make different comparisons; pin conversion/resampling/frame policy.
Development test record
Existing package/export/audio parity fixtures; no blanket identity across arbitrary codecs/streams.
Historical development records reviewed; these tests were not rerun for this reference review.Sources & basis 5 references
- T2NIST FIPS 180-4 (August 2015). Secure Hash Standard. DOI: 10.6028/NIST.FIPS.180-4.
Standard accessed; SHA family. No NIST certification of Tyree asserted.
- G4SWGDE. Best Practices for Digital Forensic Video Analysis, 18-V-001, version 2.0 (December 10, 2025), final; cited edition.
Official document accessed; workflow context.
- N3Smyly, John Woodward (2023). Investigating the Stream Hashing Function in FFmpeg. Master’s thesis, University of Colorado Denver.
Reproducibility and hash-domain controls; no claim of exact Tyree reproduction. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
- N4Kost, Susan Clarisa (2025). A Quantitative Study of Frame Extraction Methods, Tools, and Their Impact on Extracted Frame Quality from HEVC Videos. Master’s thesis, University of Colorado Denver.
Extraction/conversion test design and representation limits. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
- D8FFmpeg project. Formats Documentation, hash and streamhash muxers (rolling documentation, reviewed November 10, 2026).
Hash domain and default decoded-sample representation; inspect the pinned executable rather than assuming documentation equals installed behavior.
Follow the source.
References include research papers, standards, technical guidance and implementation documentation. Their role and access limits are stated individually.
External source links require internet access and open in a new tab. All method descriptions and reference notes on this page remain available offline. Searching above filters method records; this complete reference library remains available below.
NCMF research theses7 records
- N1Aeschliman, Austin Jon (2020). Butt-Splice Detection on Audio Recordings with Lossy Compression. Master’s thesis, University of Colorado Denver.
Method comparison, experimental design and limitations; not validation of Tyree. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
- N2Pruett, Shawn Michael (2019). An Exploratory Study of Manual Pause-Record Events in Digital Audio Recordings. Master’s thesis, University of Colorado Denver.
Alternative explanations and validation-corpus design. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
- N3Smyly, John Woodward (2023). Investigating the Stream Hashing Function in FFmpeg. Master’s thesis, University of Colorado Denver.
Reproducibility and hash-domain controls; no claim of exact Tyree reproduction. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
- N4Kost, Susan Clarisa (2025). A Quantitative Study of Frame Extraction Methods, Tools, and Their Impact on Extracted Frame Quality from HEVC Videos. Master’s thesis, University of Colorado Denver.
Extraction/conversion test design and representation limits. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
- N5Lemon, Christopher Douglas (2022). Detecting Cloned Regions in Digital Images Before and After Facebook Recompression. Master’s thesis, University of Colorado Denver.
Repeated-content test corpus and interpretation limits. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
- N6Latham, Jason (2024). Evaluation of Interpolation Methods and Their Impact on Resizing Images for Use in Digital and Multimedia Evidence (DME) Forensics—Is the Nearest Neighbor Algorithm the Best Solution in All Situations? Master’s thesis, University of Colorado Denver.
Interpolation documentation and targeted test design. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
- N7Zjalic, James (2017). A Proposed Framework for Forensic Audio Enhancement. Master’s thesis, University of Colorado Denver.
Workflow framework and outcome interpretation; no universal optimum. Full PDF locally acquired and hashed; review coverage recorded in NCMF-reading-records.json.
Research & historical context20 records
- P1Popescu, A. C., & Farid, H. (2005). Exposing Digital Forgeries in Color Filter Array Interpolated Images. IEEE Transactions on Signal Processing, 53(10), 3948–3959.
Author-hosted manuscript reviewed; CFA method background.
- P2Johnson, M. K., & Farid, H. (2005). Exposing Digital Forgeries by Detecting Inconsistencies in Lighting. ACM Multimedia and Security Workshop.
Author-hosted paper reviewed; lighting assumptions and model.
- P3O’Brien, J. F., & Farid, H. (2012). Exposing Photo Manipulation with Inconsistent Reflections. ACM Transactions on Graphics, 31(1). Appendix A.
Author manuscript reviewed; geometric constraints only. Tyree does not implement the paper’s reflection-analysis system.
- P4Christlein, V., Riess, C., Jordan, J., Riess, C., & Angelopoulou, E. (2012). An Evaluation of Popular Copy-Move Forgery Detection Approaches. IEEE TIFS, 7(6), 1841–1854. DOI: 10.1109/TIFS.2012.2218597.
Research record and abstract reviewed; evaluation context, not validation of Tyree’s hybrid.
- P5Lowe, D. G. (2004). Distinctive Image Features from Scale-Invariant Keypoints. International Journal of Computer Vision, 60(2), 91–110.
Author-hosted manuscript reviewed; SIFT and matching foundation.
- P6Muja, M., & Lowe, D. G. (2009). Fast Approximate Nearest Neighbors with Automatic Algorithm Configuration. VISAPP.
Author-hosted paper reviewed; approximate matching foundation.
- P7Fischler, M. A., & Bolles, R. C. (1981). Random Sample Consensus: A Paradigm for Model Fitting with Applications to Image Analysis and Automated Cartography. Communications of the ACM, 24(6), 381–395.
Originating institution citation checked; generic RANSAC, not app-specific acceptance rules.
- P8Richardson, W. H. (1972). Bayesian-Based Iterative Method of Image Restoration. JOSA, 62(1), 55–59. DOI: 10.1364/JOSA.62.000055.
Publisher record and abstract checked; full-text equivalence not established.
- P9Lucy, L. B. (1974). An Iterative Technique for the Rectification of Observed Distributions. Astronomical Journal, 79, 745–754.
Original paper hosted by university reviewed.
- P10Hill, N. R., & Ioup, G. E. (1976). Convergence of the van Cittert Iterative Method of Deconvolution. JOSA, 66(5), 487–489. DOI: 10.1364/JOSA.66.000487.
Publisher record/abstract checked. Convergence is conditional; this is not the original historical invention paper.
- P11Landweber, L. (1951). An Iteration Formula for Fredholm Integral Equations of the First Kind. American Journal of Mathematics, 73(3), 615–624. DOI: 10.2307/2372313.
Original bibliographic identity corroborated by the EPFL paper; publisher full text inaccessible in this review.
- P12Sage, D., et al. (2017). DeconvolutionLab2: An Open-Source Software for Deconvolution Microscopy. Methods, 115, 28–41.
Author-hosted paper reviewed; Van Cittert, Landweber, regularized inversion and Richardson–Lucy background.
- P13Wiener, N. (1949). Extrapolation, Interpolation, and Smoothing of Stationary Time Series: With Engineering Applications. MIT Press.
Historical foundation; publisher search record and EPFL bibliography checked. Book full text not obtained.
- P14Harris, F. J. (1978). On the Use of Windows for Harmonic Analysis with the Discrete Fourier Transform. Proceedings of the IEEE, 66(1), 51–83.
Original paper reviewed; windowing and spectral interpretation.
- P15Janssen, A. J. E. M., Veldhuis, R. N. J., & Vries, L. B. (1986). Adaptive Interpolation of Discrete-Time Signals That Can Be Modeled as Autoregressive Processes. IEEE TASSP, 34(2), 317–330.
Author institution record/abstract reviewed; AR interpolation family, not proof FFmpeg implements this paper exactly.
- P16Giannoulis, D., Massberg, M., & Reiss, J. D. (2012). Digital Dynamic Range Compressor Design—A Tutorial and Analysis. JAES, 60(6), 399–408.
Publisher citation checked; compressor design background.
- P17Canny, J. (1986). A Computational Approach to Edge Detection. IEEE TPAMI, 8(6), 679–698.
Original paper reviewed; family context for video edge view, not the still-image Sobel operation.
- P18Farid, H. (2006). Digital Image Ballistics from JPEG Quantization. Dartmouth Technical Report TR2006-583.
Institutional record/abstract checked. Tyree does not implement its camera/software identification database.
- P19Agarwal, S., & Farid, H. (2017). Photo Forensics from JPEG Dimples. IEEE WIFS.
Paper reviewed. Related JPEG research only; not support for Tyree’s ELA-style method.
- P20Krawetz, N. (2007). A Picture’s Worth… Digital Image Analysis and Forensics. Black Hat USA whitepaper.
Historical ELA reference located from author/conference sources. Conference whitepaper, not peer-reviewed validation; app formula is specified separately.
Standards7 records
- T1ITU-T T.81 (September 1992). Digital Compression and Coding of Continuous-Tone Still Images—Requirements and Guidelines.
Original standard accessed; JPEG structure/quantization background.
- T2NIST FIPS 180-4 (August 2015). Secure Hash Standard. DOI: 10.6028/NIST.FIPS.180-4.
Standard accessed; SHA family. No NIST certification of Tyree asserted.
- T3ITU-R BS.1770-5 (November 2023). Algorithms to Measure Audio Programme Loudness and True-Peak Audio Level.
Official current/superseded-version listing checked. App conformance not tested here.
- T4EBU R 128, version 5.0 (November 2023). Loudness Normalisation and Permitted Maximum Level of Audio Signals.
Official recommendation checked; app exposes adjustable loudness target.
- T5C2PA Technical Specification, version 2.2 (cited edition).
Cited edition accessed; do not call it latest or infer complete validator conformance.
- T6International Color Consortium. ICC.1 v4 specification landing page; existing guide cites ICC.1:2022.
Official source located; exact library conformance and display calibration not established.
- T7CIPA. Exif standards catalog.
Official catalog checked. Parser supports a subset, not claimed full support for a particular latest Exif edition.
Professional guidance5 records
- G1SWGDE. Minimum Requirements for Testing Tools Used in Digital and Multimedia Forensics, 18-Q-001, version 2.1 (March 7, 2024), final.
Testing-category guidance accessed; no endorsement/certification.
- G2SWGDE. Best Practices for Digital Audio Authentication, 15-A-001, version 1.3 (September 20, 2018), final, sections 4.5.1–4.5.4.
Waveform, spectral and first-derivative examination guidance checked. Not a calibration of Tyree thresholds.
- G3SWGDE. Best Practices for Enhancement of Digital Audio, 20-A-001, version 2.0 (August 21, 2025), final; cited edition.
Official document accessed; workflow guidance, not exact-filter validation.
- G4SWGDE. Best Practices for Digital Forensic Video Analysis, 18-V-001, version 2.0 (December 10, 2025), final; cited edition.
Official document accessed; workflow context.
- G5SWGDE. Fundamentals of Resizing Imagery and Considerations for Legal Proceedings, 22-V-001, version 1.1 (December 16, 2025), final; cited edition.
Official document accessed; resampling context.
Implementation documentation8 records
- D1FFmpeg project. Filters Documentation (rolling reference, checked November 10, 2026).
Implementation documentation. Current documentation must be reconciled with shipped build.
- D2FFmpeg project. vf_convolve.c (rolling upstream source).
Deconvolution implementation reference; not independently validated Tyree behavior.
- D3libjpeg-turbo project. src/jcparam.c (rolling upstream source).
IJG quality scaling convention, not universal quality metadata.
- D4OpenCV 4.13.0. Sobel Derivatives.
Operator reference; Tyree still-image implementation is custom.
- D5FFmpeg project. FFprobe Documentation.
Stream/frame metadata interpretation interface.
- D6MediaArea. MediaInfo official documentation/product description.
Parser implementation reference; metadata is not verified ground truth.
- D7Content Authenticity Initiative. c2patool release v0.28.6.
Release reference matched to local dependency note; executable not independently reverified here.
- D8FFmpeg project. Formats Documentation, hash and streamhash muxers (rolling documentation, reviewed November 10, 2026).
Hash domain and default decoded-sample representation; inspect the pinned executable rather than assuming documentation equals installed behavior.
Textbook foundations4 records
- B1Smith, S. W. (1997). The Scientist and Engineer’s Guide to Digital Signal Processing, chapter 24, Linear Image Processing.
Author-hosted textbook chapter reviewed; convolution, smoothing, sharpening and restoration families.
- B2Smith, S. W. (1997). The Scientist and Engineer’s Guide to Digital Signal Processing, chapter 19, Recursive Filters.
Author-hosted textbook chapter reviewed; recursive-filter foundation.
- B3Gonzalez, R. C., & Woods, R. E. (2018). Digital Image Processing, fourth edition. Pearson.
Author textbook metadata/scope checked. Full book/page-level verification not performed; use as broad background only.
- B4Smith, S. W. (1997). The Scientist and Engineer’s Guide to Digital Signal Processing, chapter 25, Special Imaging Techniques.
Author-hosted chapter accessed; image SNR and averaging context.
What this review establishes.
Documented here
- 56 method and workflow records.
- 51 reference records with source-type and access notes.
- Differences between published methods and Tyree variants.
- Existing development checks and known limitations.
Work still required
- Representative testing of custom diagnostics and thresholds.
- Version-specific links between source, shipped builds and test records.
- Independent validation where applicable.
- Confirmation of original development-source consultation history.
Product scope & citation notes
The primary source reviewed identifies itself as Tyree Forensics 0.3.0-alpha.2, a proof-of-concept/testing build, not validated for casework or evidentiary conclusions. Independent Tyree Lite, Exhibit Builder and internal Stutchman variants were also considered. A method’s presence in a private source snapshot does not establish that it is exposed in that product.
For ELA-style recompression viewing, the historical reference has been corrected to Krawetz’s 2007 conference whitepaper, alongside JPEG foundations. The JPEG-dimples paper describes a different method and appears only as related research, not support for an implemented dimples detector.
Reference lists and older notes do not, by themselves, prove which sources were consulted during original development. This is a development-method review, not an independent validation report.
GOP/frame-structure analysis remains a deferred idea. No such feature is claimed here.
