Research On Image Retrieval Based On Shape Feature

Nowadays, multimedia devices spread widely, the Internet develops rapidly, digital information refresh every day. In the face of a growing number of digital images, the user how to find what they want fast and efficiently from the database images, has become a hot problem of information retrieval. The appear of content-based image retrieval technologies has been a very good solution to this problem.Content-based image retrieval mainly uses the image itself which has the visual features such as color, shape, texture, spatial relationships, etc. Analyze,process and understand the image from the bottom to the top. Content-based image retrieval is more efficient than text-based image retrieval which uses manual annotation. Stability of Shape characteristics makes image retrieval technology based on shape features become the research hotspot and difficulty.The main work ofthis paper as follows:1. The research background, present situation of the content-based image retrieval system and the classic CBIR system are deeply analyzed.We briefly introduced the key technology of content-based image retrieval system:image preprocessing, image feature extraction and image feature matching.Among them,the technologies of image retrieval based on shape feature are described in detail, such as image segmentation, edge detection and shape feature extraction and description.2. For images that contain noise we proposed using mathematical morphological open and close operation to deal with the noise, the effect is remarkable.3. For the problem that canny operator edge detection algorithm did not detect the edge clearly, we put forward using the second generation of the fuzzy theory to select high and low threshold of canny edge detection algorithm automatically using single threshold selection technique on the image histogram. Slide membership function on the gradient histogram from the minimum to the maximum gradient value.Calculate ultrafuzziness based on formula.The center of the membership function of ultrafuzziness corresponds to the optimum threshold of image gradient histogram.Then use algorithm to calculate the two thresholds of Canny operator.This method minimizes fuzzy uncertainty during threshold selection.4. The proposed algorithm was applied to develop a simple image retrieval system to make trademark images retrieval experiments, retrieval results are accurate.It can accurately find the most similar images to the key figure, and display the five highest similar images.The edge detection result is clear.