The Research Of Key Technology Of Content-based Image Retrieval

In recent years, driven by the rapid development of computer network and multimedia technologies, the multimedia data which is represented by image and video are growing at a remarkable rate. And people eager to be able to search the desired images from lager numbers of images quickly. However, the early text-based image retrieval technology has been unable to meet people’s needs for fast retrieval. In order to solve this problem, the content-based image retrieval technology was born.CBIR got widespread attention and a series of techniques about it were developed from scholars at home and abroad since the 90 s of last century. On the basis of the research achievements of many experts in this field in recent years, several key technologies of CBIR were discussed and image retrieval based on color and shape feature were focused on in this paper. The focus of this paper is to improve the algorithms of color-based image retrieval and put forward a new algorithm of shape-based image retrieval. What his pager studied are as follows:Firstly, after the general framework structure of CBIR system is illustrated, several key technologies of CBIR is studied and analyzed deeply, including the extraction and expression of underlying feature of image, the theorem of similarity measure and its algorithms used commonly, a brief description of several evaluation criterion of image retrieval.Secondly, improved color histogram algorithms which are based on weighted block and dominant color were presented. For the weighted block color histogram, dividing images into blocks could partly solve the problem of losing spatial information which is caused by global color histogram. And weights assigned to each block could take the position of target object in the image into account. In general, the block with the target object should be given a higher weight. For the dominant color histogram, the colors which hold a large proportion are extracted as the color information of image. The experiments show that both the improved algorithm is better than traditional color based image retrieval algorithm.Lastly, a new shape-based image retrieval algorithm was presented. Smoothness and roughness are used to describe shape feature of image, and the corresponding quantization algorithms were presented. The technologies of image enhancement and image segmentation were studied and analyzed deeply to extract shape features more easily and accurately. And shape features extraction algorithms used commonly were introducted briefly. At last, the experiment together with the weighted block color histogram-based image retrieval algorithm gets a better result.