Fake Colorized Image Detection Using Color Features and LBP-Encoded Gray-Level Co-occurrence Matrix

This paper proposes a Fake Colorized Image Detection (FCID) method that integrates CHMC color features (color histograms, color moments, and color contrast) with a Gray-Level Co-occurrence Matrix (GLCM) specifically adapted to the Local Binary Pattern (LBP) encoding framework. First, the input image is transformed from the RGB color space to the CIELAB color space. Then, CHMC color features and GLCM-based features in the LBP encoding domain are extracted independently. The CHMC color features comprise color histograms, color moments, and color contrast, designed to capture distributional variations in luminance and chrominance resulting from recoloring manipulations. For the GLCM computed within the LBP map, Non-Rotationally Invariant Uniform Local Binary Patterns (NRI-ULBP) histograms are initially extracted from each CIELAB channel. Subsequently, the GLCM is constructed from the LBP map to model spatial dependencies among texture-encoded patterns. Experimental results indicate that the proposed method consistently surpasses the compared methods in terms of detection performance. It exhibits stable accuracy across varying degrees of image compression, highlighting its robustness to common post-processing distortions. Furthermore, the results confirm that the CHMC color features and the GLCM derived from the LBP-encoded domain are both effective and complementary for the task of FCID.