Local energy oriented pattern for image indexing and retrieval

Abstract

A novel image indexing algorithm for Content Based Image Retrieval (CBIR) using Local Energy Oriented Patterns (LEOP) is proposed in this paper. LEOP encodes pixel level energy orientations to find minute spatial features of an image whereas existing methods use neighborhood relationship. LEOP maps four pixel progression orientations to find top two maximum energy changes for each reference pixel in the image i.e. for each reference 3 3 grid, two more 3 3 grids out of four pixel progression are extracted. Finally, LEOP encodes the relationship among pixels of three 3 3 local grids extracted. LEOP is applied on four different image databases named MESSIDOR, VIA/I-ELCAP, COREL and ImageNet Database using traditional CBIR framework. To test the robustness of proposed feature descriptor the experiment is extended to a learning based CBIR approach on COREL database. The LEOP outperformed state-of-the-art methods in both traditional as well as learning environments and hence it is a strong descriptor.