An Effective Content Descriptor Based on Block-Edge-Pattern in DCT Bomain for Image Retrieval

摘要

A robust and effective content descriptor based on block edge patterns is described in this paper. This content descriptor is constructed by an edge-block histogram with three patterns of horizontal-edge, vertical-edge and no-edge, which are directly extracted in DCT domain. It is shown that the proposed features generate very low computing cost and high processing speed for compressed images or videos. The proposed method is robust to a range of orientation changes, which are frequently encountered during transmission, processing and storages. Experimental results show the superiority and robustness in comparison with existing descriptors.