Research On Application Techniques Of Content-based Image Retrieval(CBIR)

With the rapid development of database system and computer vision, content-based image retrieval technique has become a very hot point of study. In this paper, we deal with this subject in three different basic features respectively. We mainly focus on algorithm studies and introductions of methods for color, shape and texture feature extracting and performance assessment, which are widely used in CBIR systems. And we manage to conclude and summarize algorithms for each feature, then select several color and shape extracting algorithms to carry out an experiment.As for color-based image retrieval, we propose a relatively improved method for color feature extraction to dynamically decide the color quantization according to the inherent content of the given image. Then we adopt color correlogram to describe color features. For shape-based image retrieval, at first, we use Log operator to detect edge and then employed Fourier descriptor based on two-dimensional polar Fourier transform for shape description, which captures not only the boundary features but also the region features. And thus the method is application independent and robust. As for texture feature, we haven't done deep study but present some introductions for several of the typical methods. Experimental results show that we have balanced the efficiencies and performances of the retrieval system, and got better retrieval results.