Image Retrieval using One-Dimensional Color Histogram Created with Entropy

Abstract

Image histograms are frequently used as a feature vector in content-based image retrieval (CBIR). The related methodology involves processing of a single channel histogram on gray level images while histograms of three channels must be processed in color images. Subsequently, there are two ways to process histograms of color images. In the first approach, the length of feature vector is extended by adding histogram data of each channel to create new feature vector. However, this kind of solution increases computational time and complexity. Second solution is to combine the histogram data obtained from each channel to establish a feature vector. In this study, a novel image retrieval approach, which uses a cluster-based one-dimensional histogram (ODH) for color images has been developed. Initially, multiple thresholds (MT) for each channel were calculated by means of Kapur entropy method. Then, the RGB color space was subdivided into sub-cubes or prisms. The numbers of pixels in each cluster and cluster index or class label have been used to construct a cluster-based one-dimensional histogram. Finally, image retrieval process has been implemented by using the one-dimensional color histogram (ODH) of images in database and query.

Keywords

• entropy
• feature extraction
• histograms
• image retrieval
• vector quantization