Color and Shape Based Image Representation and Retrieval

A novel and efficient scheme for feature extracting，indexing，and retrieving color image is presented．To extract color feature，a simple invariant labeling function is defined to measure the hue of each pixel，and hue histogram is generated which represents the distribution of hue values in an image．To describe the relative position and orientation of objects within the image，hue edges in the image，computed using Canny approach，are processed by Radon transformation to generate the spatial histogram which is used as shape representation．To address the issue of image matching，we define two global distance measures based on the accumulative distance and vector-based distance to compare the similarity of two images．In addition，some key problems such as the combination of different subfeatures and normalization of distance measures are also discussed in detail．Experimental results indicate that our approach provides perceptually appealing results and high retrieval rate．