Hierarchical visual perception and two-dimensional compressive sensing for effective content-based color image retrieval

Content-based image retrieval (CBIR) has been an active research theme in the computer vision community for over two decades. While the field is relatively mature, significant research is still required in this area to develop solutions for practical applications. One reason that practical solutions have not yet been realized could be due to a limited understanding of the cognitive aspects of the human vision system. Inspired by three cognitive properties of human vision, namely, hierarchical structuring, color perception and embedded compressive sensing, a new CBIR approach is proposed. In the proposed approach, the Hue, Saturation and Value (HSV) color model and the Similar Gray Level Co-occurrence Matrix (SGLCM) texture descriptors are used to generate elementary features. These features then form a hierarchical representation of the data to which a two-dimensional compressive sensing (2D CS) feature mining algorithm is applied. Finally, a weighted feature matching method is used to perform image retrieval. We present a comprehensive set of results of applying our proposed Hierarchical Visual Perception Enabled 2D CS approach using publicly available datasets and demonstrate the efficacy of our techniques when compared with other recently published, state-of-the-art approaches.