Scalable Face Image Retrieval with Dynamic Allocation of Attribute Weights Using Attribute: Enhanced Sparse Codewords

Abstract

Large-scale content-based face image retrieval is an enabling technology for many emerging applications. It utilizes automatically detected human attributes that contain semantic cues of the face photos to improve content based face retrieval by constructing semantic codeword's for efficient large-scale face retrieval. Here attributes are dynamically decided by depending upon its importance and then further exploit the contextual relationships between them. Then by leveraging human attributes in a scalable and systematic framework, two orthogonal methods named attribute-enhanced sparse coding and attribute embedded inverted indexing is proposed to improve the face retrieval in the offline and online stages. Experimental result shows better result when compare with existing one.