Research On Image Retrieval Based On Salient Closed Edge

In the traditional content-based image retrieval technology, the most commonly usedimage features include color, texture and shape. Edge, as a very important and useful visualfeature, only uses limited pixels to describe the most visual information. In order to improvethe retrieval precision of the contect based image reieval, this paper presents a method ofextracting the closed salient edge based on visual attention mechanism. The method firstcombines the saliency map received by using human visual attention mechanism and theoriginal edge image to obtain significant edge, and then connects the boundaries betweenevery salient edge to obtain the closed salient edge. Finally, the method achieves the imageretrieval effectively according to the the gradient histogram and direction histogram of theedge to. This paper mainly studies edge detection method, method for salient edge extractionand salient closed edge extraction, image retrieval based on salient closed edge.The main work and conclusions are as follows:(1) Analyzing and comparing several edge detection methods and chosing the mostsuitable one for this paper. Because of the complexity of the image itself, although peoplehave been committed to the realization of effective methods on edge detection and differentpurposes to use the edge information, there is not a universal algorithm for edge detection andstandard for evaluation. In order to obtain the edge detection method appropriate for thispaper which is convenient for the subsequent algorithm processing,, different edge detectionmethods including Sobel, Prewitt., Roberts, LOG, Canny and PM algorithm have beenanalyzed and tested. After a variety of analysis, comparison and evaluation on edge detectionalgorithms, PM algorithm has been chosed for edge detection to get the original edge image.(2) Combining the saliency map and the original edge image to obtain the salient edge.After researching Itti's model of visual attention mechanism, a simplification of the method toobtain the saliency map was proposed; then, achieving the extraction of the salient edge bycombining the saliency map and the original edge image. Through comprehensive use ofsaliency value and the length of edges to evaluate the salient edge, edges which do not meetthe requirements of the proposed method for image retrival are deleted. The experimentproves that using the proposed method in this paper, satisfacted salient edge images can beachieved.(3) Research and design an effective algorithm for edge linking and obtain the closedsalient edge. There are a lot of discontinuous edges in the salient edge image, especially some discrete short edges and isolated points which can not represent the useful information of theoriginal image. Moreover, the discontinuous salient edges influence the result of the imageretrieval. Therefore, after salient edge extraction, edge linking is a necessary for connectingthe discontinuous salient edges in order to obtain the closed salient edge which is animportant preparation of image retrieval. By analysing various mainstream edge linkingalgorithms, a suitable method is designed for the edge linking to realize closed salient edgeextraction. By using5parameters which are the number of original edges (OE), salient edges(SE), closed salient edges (CSE), self-closed edges (SCE) and the final correct connectingrate (CCR), the effectiveness of the propsed method for edge linking can be evaluated.(4) Researching image retrieval method based on the closed salient edge and developinga simple image retrival system. The method mainly uses the edge gradient and derectionhistograms as features for image retrieval. As for some images containing the more complexnatural scenes, the color features can be combined with; developing a simple image retrievalsystem to verify the effectiveness of the proposed method and improve the speed andaccuracy of image retrieval.