The Research And Application Of Image Retrieval System Based On Comprehensive Feature

With the rapid development of multimedia technology and the growing emergence of massive image library, how to rapidly and accurately retrieve the related image from a large database has become an urgent problem. The traditional content-based image retrieval method is based on single feature, the method is biased and lack of precision for image retrieval. In order to fully describe the image content attribute, always needs comprehensive variety of features to retrieve. This paper mainly studied the color and shape features, and comprehensive color and shape features to retrieval, the main research content is as follows:1. In-depth analysis and research of some related technologies on CBIR in the field, based on the distribution characteristics of image color space put forward an image retrieval method based on HBCH index. On account of the distribution characteristics of image color space present a image retrieval method based on HBCH index. The method firstly use the equal-area circular dividing method to block process image and the extracte features, this way can associate the image color space, will be welled to reflect the position of image color space.2. Starting from the image shape features and on account of images interest points, proposed a Hu rectangular shape feature image retrieval.method based on a uniform Harris interest point. This method firstly adopt the interest point extraction method based on the uniform Harris operator to extract the uniform Harris interest point, then use equal-area circular dividing method to divide the interest points into four regions, and calculate the Hu rectangular shape feature of each region, then obtain the shape feature vector SF of images.3. Based on the study of color and shape features, synthetically weighted the improved color and shape features, and then retrieve image based on comprehensive features.This paper adopted the Microsoft Visual C++6.0development tool to build a algorithm validation system based on comprehensive features. Experimental data show that the two image retrieval methods which one based on HBCH index and the other one use uniform Hu rectangular shape features based on Harris interest points has the same result of retrieval precision. The result of feature information extracted by the second method with rich content, small amount of data, applied to retrieve quickly and efficiently, while the result of feature information extracted by the first method with highest precision and more accord with people of image intuitive feeling.