Format-Controlled Multi-Scale JPEG Compression Response Analysis for Image-Level Forgery Screening

2026-07-07 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy · Depth: Expert, quick

Summary

A new lightweight, interpretable feature engineering pipeline is proposed for image-level forgery screening, operating solely on CPU computation with gradient boosted trees. This method integrates multi-scale Error Level Analysis (ELA) across seven JPEG quality levels, novel cross-quality ELA ratio features for double-compression artifact detection, augmented by spatial entropy, FFT energy bands, edge density, SRM residuals, and DCT blockiness, forming a 405-dimensional feature vector. Addressing a format confound in CASIA v2.0, the pipeline was rigorously evaluated on a JPEG-only subset of 9,501 images. It achieved an AUC of 0.990 [95% CI: 0.988--0.991] and an F1 score of 0.905 using 5-fold stratified cross-validation. Even with a conservative source-aware group split, AUC remained 0.976. Ablation studies confirm multi-scale ELA's significant contribution (+0.180 AUC gain).

Key takeaway

For AI Security Engineers or digital forensics professionals needing efficient image forgery screening, this CPU-only pipeline offers a robust solution. You can achieve high accuracy (AUC 0.990) without GPU acceleration, making it suitable for resource-constrained deployments. Focus on multi-scale ELA and cross-quality ratios to detect compression-history inconsistencies, ensuring your screening methods are resilient against format-based confounds.

Key insights

Lightweight CPU-based feature engineering with multi-scale ELA effectively screens image forgeries by detecting compression-history inconsistencies.

Principles

• Multi-scale ELA significantly improves forgery detection.
• Cross-quality ELA ratios identify double-compression artifacts.
• Rigorous format-controlled evaluation prevents confounders.

Method

The method computes multi-scale ELA at seven JPEG quality levels, cross-quality ELA ratios, spatial entropy, FFT energy bands, edge density, SRM residuals, and DCT blockiness to form a 405-dimensional feature vector for gradient boosted trees.

In practice

• Use multi-scale ELA for robust forgery screening.
• Combine ELA with other image features.
• Evaluate on format-controlled datasets.

Topics

• Image Forgery Detection
• Error Level Analysis
• JPEG Compression Artifacts
• Digital Forensics
• Feature Engineering
• CPU-only Inference

Best for: Computer Vision Engineer, AI Scientist, AI Security Engineer, Research Scientist

Related on AIssential

• How Fragile Are Training-Free AI-Generated Image Detectors? A Controlled Audit of Score Direction, Preprocessing, and Compression — Training-free AI-generated image detectors exhibit significant fragility, highly sensitive to implementation details, hyperparameters, and dataset biases.
• Multi-Frequency Fusion for Robust Video Face Forgery Detection — Lightweight fusion of handcrafted cues can significantly improve face forgery detection accuracy with minimal model overhead.
• Tool-IQA: Augmenting Image Quality Assessment with Simple Tools — Tool-IQA enhances VLM-based image quality assessment by integrating dynamic inspection tools like Magnifier and Gamma Corrector.
• From Pixels to DNA: Why the Future of Compression Is About Every Kind of Data — Data compression is evolving into a foundational technology for all digital data, shifting from human-centric media to machine-centric semantics.
• Order within Chaos: Capturing Intrinsic Energy Anomalies for AI-Manipulated Image Forgery Localization — AI image generation creates a detectable "statistical energy gap" due to suppressed high-frequency variance, enabling forgery localization.
• When Recovery Matters: The Blind Spot of Surrogate Privacy in MLLM Editing — Surrogate-based MLLM image editing requires explicit recovery mechanisms to transfer edits back to private source images.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.