Research On Image Retrieval Technology Based On Region Features

With rapid advances in internet and multimedia computing technologies, how to rightly find the required images in lots of images is becoming an important problem of image retrival. Several key techniques in region-based image retrieval are studied in this paper. Major work and its creativity are listed as follows.First, an image segment algorithm based on the integration of region information and edge detection is presented. This algorithm translates the color image into the gray image, and makes region segment of gray image using ostu's segment method to extract the closed and connected image outline and compute the best threshold value which is used in the canny algorithm as the high threshold value. Then, the edge image is extracted using the canny algorithm. Finally, the segment result is obtained by interating the region's outline image and the edge image. The experiments show that this algorithm not only achieves good image segment, but also filters most detailed marginal point which is unnecessary and irregular in order to get the clear image outline.Second, an image retrieval method based on integrated features of objection region is put forward. This method first extracts the possible objection regions using the proposed image segment algorithm. After the objection region is confirmed by the user, the color features with 36 dimensinos, the textures feature with 4 dimensions and the shape feature with 7 dimensions are extracted. Then, the procedure of region match is carried out. Finally, the retrieval result is listed to user by the descending order of similarity. The experiments show that, in terms of precision-recall, this method not only is better than other methods which use the only feature, but also achieves better performance than the method which uses the global integrated features.Third, an image retrieval system is developed. By using the software of Visual C++6.0, a minitype image retrieval systeme is constructed. And the proposed image retrieval method based on the objection region is verified.