A Salient Object-Based Image Retrieval Using Shape and Color Features

Abstract

Shuxian Huang, Wenbing Chen Nanjing University of Information Science and Technology, Nanjing 210044, China (Received January 17 2018, accepted July 05 2018) In \u00a0this \u00a0paper, \u00a0a \u00a0salient \u00a0object-based \u00a0image \u00a0retrieval \u00a0method \u00a0(SOBIR) \u00a0is \u00a0presented, \u00a0which \u00a0linearly combines the shape and colour features of the salient objects contained in target and candidate images respectively to carry out content-based image retrieval (CBIR). The framework of the proposed method is carried out as follows: first, the mean shift and region growing algorithms are used to segment an input image into many regions; secondly, based on these regional contrasts the saliency map, the binary image, and the salient object image are extracted respectively; thirdly, \u00a0the \u00a0shape \u00a0representation \u00a0of \u00a0the \u00a0salient \u00a0object \u00a0is \u00a0extracted \u00a0from \u00a0the \u00a0binary \u00a0image \u00a0using \u00a0an \u00a0improved \u00a0polar Fourier Descriptor method, meanwhile the salient object contained in the input image is converted into a representation of its histogram in the L\u2217a\u2217b\u2217 colour space; Finally, the similarity between the two salient objects contained in the target and candidate images is defined by linearly combining both the shape and colour representations. Experimental results \u00a0show \u00a0that, \u00a0compared \u00a0to \u00a0the \u00a0latest \u00a0two \u00a0CBIR \u00a0methods, \u00a0the \u00a0proposed \u00a0SOBIR \u00a0method \u00a0exhibits \u00a0an \u00a0excellent performance in precision, recall, flexibility and efficiency.