ResAIKit
Research Integrity Toolkit
Back to the encyclopedia
G1-imgImage forensicsChart AnalysisLayer 1 (Deterministic)

Vectorial vs Raster

Determines whether a chart was created by software (clean, sharp lines) or captured as a photograph/screenshot, which can reveal whether the chart is authentic or reconstructed.

Technical description

Analyzes the image to distinguish between vector-origin charts (exported from plotting software as rasterized vector graphics) and photographic captures. Examines: resolution consistency across the image, presence/absence of JPEG compression artifacts around sharp edges, anti-aliasing quality, line straightness and regularity, and text rendering quality. Vector-origin charts have pixel-perfect lines and uniform backgrounds, while photographed charts show optical imperfections.

How it works

Layer 1 (deterministic): Analyzes edge quality for line straightness and anti-aliasing patterns. Checks background uniformity for camera noise. Detects JPEG artifacts around text and lines. Measures resolution consistency. Evaluates text rendering quality using OCR confidence. Reports whether the chart appears vector-generated, screenshot, or photographic.

Why this matters

Understanding whether a chart is a clean software export versus a photograph or screenshot provides context for interpreting other forensic results. A chart that appears to be freshly generated from software might be authentic or might have been recreated to present fabricated data. Charts photographed from screens may indicate data provenance issues.

Score thresholds

0-1
Clean vector-origin chart consistent with direct software export
2-3
Mixed characteristics, possibly re-exported or screenshotted
4-5
Significant inconsistencies suggesting recreation or heavy manipulation

Limitations

High-resolution screenshots of vector charts can look identical to direct exports. Some plotting software produces lower-quality output. The distinction between 'recreated' and 'original' charts requires additional context.

References

  1. Hu Y, Zhao C, Mo C, Liu H, Li X. (2026). SciFigDetect: a benchmark for AI-generated scientific figure detection. arXiv:2604.08211
  2. Lin X, Tang W, Wang H, Liu Y, Ju Y, Wang S, Yu Z. (2024). Exposing image splicing traces in scientific publications via uncertainty-guided refinement. Patterns (Cell Press)
  3. Kwon MJ, Nam SH, Yu IJ, Lee HK, Kim C. (2022). Learning JPEG compression artifacts for image manipulation detection and localization. International Journal of Computer Vision
  4. Liang M, Qu Y, Jiang Y, Backes M, Zhang Y. (2025). From Evidence to Verdict: An Agent-Based Forensic Framework for AI-Generated Image Detection. arXiv:2511.00181