Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Handcrafted Feature Fusion for Reliable Detection of AI-Generated Images

Syed Mehedi Hasan Nirob, Moqsadur Rahman, Shamim Ehsan, Summit Haque

TL;DR

The paper tackles the detection of AI-generated images by evaluating a broad set of handcrafted descriptors on CIFAKE across three feature configurations (baseline, advanced, mixed) using seven classifiers. It finds that performance improves monotonically with richer feature sets and that gradient-boosted trees, especially LightGBM, offer the best discrimination and calibration (PR-$AUC$ up to $0.9879$ and Brier score as low as $0.0414$). The work provides a comprehensive benchmark demonstrating that carefully engineered features retain strong detection capability, offering interpretability and efficiency advantages in constrained or forensics-critical settings. It also highlights limitations such as dataset scope and resolution, and suggests future exploration of hybrid handcrafted-deep approaches and higher-resolution data for broader generalization.

Abstract

The rapid progress of generative models has enabled the creation of highly realistic synthetic images, raising concerns about authenticity and trust in digital media. Detecting such fake content reliably is an urgent challenge. While deep learning approaches dominate current literature, handcrafted features remain attractive for their interpretability, efficiency, and generalizability. In this paper, we conduct a systematic evaluation of handcrafted descriptors, including raw pixels, color histograms, Discrete Cosine Transform (DCT), Histogram of Oriented Gradients (HOG), Local Binary Patterns (LBP), Gray-Level Co-occurrence Matrix (GLCM), and wavelet features, on the CIFAKE dataset of real versus synthetic images. Using 50,000 training and 10,000 test samples, we benchmark seven classifiers ranging from Logistic Regression to advanced gradient-boosted ensembles (LightGBM, XGBoost, CatBoost). Results demonstrate that LightGBM consistently outperforms alternatives, achieving PR-AUC 0.9879, ROC-AUC 0.9878, F1 0.9447, and a Brier score of 0.0414 with mixed features, representing strong gains in calibration and discrimination over simpler descriptors. Across three configurations (baseline, advanced, mixed), performance improves monotonically, confirming that combining diverse handcrafted features yields substantial benefit. These findings highlight the continued relevance of carefully engineered features and ensemble learning for detecting synthetic images, particularly in contexts where interpretability and computational efficiency are critical.

Handcrafted Feature Fusion for Reliable Detection of AI-Generated Images

TL;DR

The paper tackles the detection of AI-generated images by evaluating a broad set of handcrafted descriptors on CIFAKE across three feature configurations (baseline, advanced, mixed) using seven classifiers. It finds that performance improves monotonically with richer feature sets and that gradient-boosted trees, especially LightGBM, offer the best discrimination and calibration (PR- up to and Brier score as low as ). The work provides a comprehensive benchmark demonstrating that carefully engineered features retain strong detection capability, offering interpretability and efficiency advantages in constrained or forensics-critical settings. It also highlights limitations such as dataset scope and resolution, and suggests future exploration of hybrid handcrafted-deep approaches and higher-resolution data for broader generalization.

Abstract

The rapid progress of generative models has enabled the creation of highly realistic synthetic images, raising concerns about authenticity and trust in digital media. Detecting such fake content reliably is an urgent challenge. While deep learning approaches dominate current literature, handcrafted features remain attractive for their interpretability, efficiency, and generalizability. In this paper, we conduct a systematic evaluation of handcrafted descriptors, including raw pixels, color histograms, Discrete Cosine Transform (DCT), Histogram of Oriented Gradients (HOG), Local Binary Patterns (LBP), Gray-Level Co-occurrence Matrix (GLCM), and wavelet features, on the CIFAKE dataset of real versus synthetic images. Using 50,000 training and 10,000 test samples, we benchmark seven classifiers ranging from Logistic Regression to advanced gradient-boosted ensembles (LightGBM, XGBoost, CatBoost). Results demonstrate that LightGBM consistently outperforms alternatives, achieving PR-AUC 0.9879, ROC-AUC 0.9878, F1 0.9447, and a Brier score of 0.0414 with mixed features, representing strong gains in calibration and discrimination over simpler descriptors. Across three configurations (baseline, advanced, mixed), performance improves monotonically, confirming that combining diverse handcrafted features yields substantial benefit. These findings highlight the continued relevance of carefully engineered features and ensemble learning for detecting synthetic images, particularly in contexts where interpretability and computational efficiency are critical.
Paper Structure (20 sections, 13 equations, 3 figures, 3 tables)

This paper contains 20 sections, 13 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Deep Learning Approaches
  4. Handcrafted and Hybrid Approaches
  5. Positioning of This Work
  6. Problem Description
  7. Methodology
  8. Dataset and Split
  9. Feature Extraction
  10. Classifiers
  11. Training and Thresholding
  12. Evaluation Metrics
  13. Implementation Details and Reproducibility
  14. Results
  15. Baseline descriptors
  16. ...and 5 more sections

Figures (3)

  • Figure 1: Sample images from the CIFAKE dataset. Top three rows: Real images. Bottom three rows: Synthetic or AI-generated images.
  • Figure 2: Balanced accuracy of different models across three feature sets (Baseline, Advanced, Mixed). Accuracy improves consistently with richer feature sets.
  • Figure 3: LightGBM performance across different feature sets. All metrics improve monotonically from baseline to mixed descriptors.