DefinePK

Abstract

This study presents to construct special networks from images and then apply their topological properties to image retrieval. Each input color image is divided into three separate gray-level images in the RGB space. For each gray-level image, we view the 256 gray-levels as nodes and construct the Horizontal Gray-level Co-occurrence Network (HGCN) and Vertical Gray-level Co-occurrence Network (VGCN) by counting the number of horizontal and vertical occurrences for each possible gray-level pair. Based on the obtained six directed weighted networks HGCN_R/G/B and VGCN_R/G/B, we extract their topological features including in-degrees, out-degrees, in-strengths and out-strengths for image retrieval. Simulation results demonstrate the superiority of our features to some existing features in terms of P-R curve.