Retrieval of occluded images using DCT phase and region merging

In this paper we present an efficient method for Content Based Image Retrieval (CBIR) of occluded images using DCT-phase. The proposed method utilizes a novel correlation metric for ternary-valued DCT-phase, as well as a region merging method to reconstruct the non-occluded regions in the retrieved image. The proposed image retrieval method showed good performance when tested with different datasets containing reference images, occluded images, fused images and images with different JPEG compression ratios. Experimental evaluation also showed that the proposed image retrieval method performs better than current state of the art DCT-phase based image retrieval methods while retrieving not only occluded images but also reference images, fused images and images with different JPEG compression ratios.