FORENSIC SKETCH DETECTION BASED ON ANFIS AND MULTIVIEW LEARNING

Abstract

 In forensic science, accurately linking a hand-drawn sketch to a real photo remains a challenging task due to the sensory gap between modalities and artistic exaggerations in sketches. Conventional techniques—whether handcrafted feature extraction or deep learning transformations—often struggle to generalize across varied facial structures. This paper introduces a sketch recognition framework that integrates Multiview Learning with an ANFIS for dynamic feature fusion. The method employs a GAN to generate photo-like images from sketches, reducing cross-modal disparity. Four complementary descriptors—HOG, ORB, LBP, and DoG—are then extracted as multiple views. Finally, ANFIS adaptively assigns feature weights to optimize the fusion process. Experimental evaluation on the CUFS dataset and the AR face dataset demonstrates that the proposed system surpasses state-of-the-art approaches, achieving 98.2% Rank-1 accuracy on CUFS and 97.65% on AR, with 100% precision at Rank-3 and Rank-5. Further analysis using SSIM, FSIM, and VIF metrics validates the strength of ANFIS-driven fusion. Unlike traditional fuzzy systems relying on fixed rules, ANFIS dynamically tunes feature selection, enhancing robustness and generalization. The results establish a new benchmark in forensic sketch recognition, highlighting the practical significance of combining Multiview Learning with adaptive fusion strategies.