Hybrid SIFT-DCT Approach for Face Matching of BuddhaStatues: Addressing Negative Similarity Metrics with SHashing

Abstract

Art and cultural heritage rely on image processing techniques forpreservation and analysis. A key challenge in this study is accuratelydetecting highly similar Buddha faces despite variations in lighting, rotation,and minor facial differences. This paper proposes a Content-Based ImageRetrieval (CBIR) framework that integrates Discrete Cosine Transform(DCT) and Scale-Invariant Feature Transform (SIFT) to enhance face-matching accuracy. The system is tested on a database of Buddha imagescharacterized by intricate textures and fine details, where DCT extractsglobal texture representations while SIFT captures localized structuralfeatures. Experimental results demonstrate that while DCT effectivelyencodes global texture characteristics, SIFT enhances local feature detectionbut struggles to differentiate between Buddha faces with extremely highsimilarity. One of the primary challenges encountered was the instability intexture similarity computation, where Chi-Square Similarity produced a-39.44% value for certain statues due to noise, artifacts, and lightinginconsistencies. These findings highlight the importance of robustpreprocessing techniques and refined similarity metrics to improve retrievalconsistency. Overall, the hybrid DCT-SIFT approach improves the accuracyand robustness of CBIR systems in historical artifact datasets. Futureresearch should focus on optimizing preprocessing steps, integratingadaptive feature selection, and exploring more stable similaritymeasurement techniques to further enhance retrieval performance.