ATTRIBUTE-ENHANCED SPARSE CODEWORDS AND INVERTED INDEXING FOR SCALABLE FACE IMAGE RETRIEVAL

Abstract — Social networks have become popular due to its photo sharing facilities. People are interested to explore contents that contain images. Since internet has become a part of life people are interested in uploading images in it. Hence with the exponentially growing photos, large-scale content-based face image retrieval is a facilitating technology for many emerging applications. In this paper, our aim is to utilize automatically detected human attributes which contain semantic cues of the face photos to improve content based face retrieval by constructing semantic cues for efficient image retrieval. This technique is coupled with relevance ranking technique to enhance the efficiency further. Two orthogonal methods named attribute-enhanced sparse coding and attribute embedded inverted indexing are proposed to improve the image retrieval in both offline and online stages. Relevance ranking when added with these methods show performance improvement to greater extent