Research On Visual Mechanism Based Image Saliency Region Detection And Retrieval Algorithm

The number of the digital image is increasing unexpectedly as the information technology is improving fast. How to process the image precisely and fast by computer has been the one of the most important researches studied by the researchers. The saliency detection is one of the most important steps in the image processing. Introducing the visual mechanism into computer vision, then we can mimic visual system to process and analyze images efficiently, which has been one of the hot topic in recent years.Based on systematic research of basic theories, technologies and some representative of the image saliency detection, in order to solve the problems of poor results and losing information, a new algorithm of visual mechanism based saliency detection and extraction is presented. A new algorithm of image retrieval can also be presented by applying the visual mechanism into image retrieval. The contribution and relevant work in this paper are as follows:Firstly, a new algorithm of Absorbing Markov chain based image saliency detection is presented. We analyze the theories and algorithms of image saliency detection. Aimed to the poor results of the detection, we use the SLIC super pixels to segment images and the absorbing Markov chain to saliency detect by computing absorbing time. The results of saliency detection can meet the needs, and efficiency can be raised 40% compared to Itti algorithm, the quality can also be improved obviously.Secondly, a new algorithm of visual mechanism based saliency region extraction is presented. We analyze the results of saliency detection, and then present a mapping algorithm of saliency region extraction, which designed by using XYZ color space based the two-dimensional function, to solve the problem that images will lose some important information of saliency region and cannot be used to further process because of the poor results of the saliency detection. The experiments show that our algorithm can extract the regions of focusing precisely.Thirdly, a new algorithm of regional color histogram based image retrieval which separated by visual mechanism is presented. Based on the research of algorithms of image retrieval, a new algorithm is presented to solve the problems that the results of image retrieval can't match human's focusing. Introducing the visual mechanism to image retrieval, we use the images whose region have been extracted to replace the whole images, compute the color moments, then retrieve images by using improved algorithm of color histogram. The results of experiments show that our algorithm can improve accuracy of image retrieval with 20% compared to traditional algorithm.Fourthly, based on the algorithms presented above, a software frame VMIR of visual mechanism based image retrieval is designed. It can detect the regions of saliency, extract the regions of saliency and retrieve the images that regions of saliency have been extracted. The tests of image retrieval show it can meet the demands of image retrieval, and it is a friendly frame.