Three online tools for detecting figure manipulation, image duplication and paper-mill patterns in published articles — moving beyond hand-crafted SIFT features to modern, learning-based pipelines.
A multimodal pipeline for screening the literature: it parses PDFs into text and figure metadata, learns joint embeddings, and flags near-duplicate articles, reused images and paper-mill signatures, built from retraction records as train/test data.
Open platformImage forensics that detects manipulation and shows where it happened. Pixel reads compression, noise and resampling traces; FakeScope localizes copy-move, splicing and generative edits, returning heatmaps and marked regions for every suspect figure.
Open platformThe SILA reference system from SciInt, reproduced here as a forensic baseline. It segments figure panels, ranks visually similar regions and traces image provenance across a paper — the yardstick newer embedding-based detectors are measured against.
Open platformPublished PDFs are processed into structured metadata: clean text, segmented figures and panels.
Text and images are encoded with modern embeddings, replacing legacy SIFT descriptors.
Similarity search across the corpus surfaces duplicates, reuse and manipulated regions.
Text and image evidence converge in a multimodal score that flags likely integrity issues.
Prof. Dr. Carlos Henrique da Silva Santos
Scientific-image forensics & computational integrity
IFSP — Campus Itapetininga · GTAC