Study On Extraction Methods Of Image Interesting Regions

Currently, image is one of the main forms to express multi-media information. It has been an urgent problem to be solved to extract the object or region of interest in image retrieval. Conventional extract methods of region of interest often focus on man-machine interaction or detecting the salient point of image. There are still some shortcomings for this method in efficiency and accuracy. In this essay, we present an improved Itti extraction method of region of interest based on biological vision principle and wavelet. At the same time, a region of interest retrieval system based on matlab has been developed to validate this method.The main content of this research are as follows:Firstly, based on the principles of biological vision attention mechanism and wavelet, we proposed a improved Itti algorithm to extract the region of interest. First, we extract the color, brightness and direction component of image , and then construct the multiscale image pyramids of every component by using wavelet multi-resolution decomposition. Then, we process the multiscale image pyramid by using the Center-Surround operator and get the three conspicuity map of the three components.Secondly, improve the construction method of the end saliency map. In order to inhibit the less salient area and enhance the salient area, we dealing with conspicuity map of the three component by using iteration method, and then get three saliency maps. Last, we combine the three saliency maps into a synthesize saliency map by using maximum method. Thirdly, the use of winner takes all (WTA) network and the region growing method generated a saliency map for processing. At first, the use of WTA network access to the most significant point, and then the use of inhibition of return mechanism to suppress the point of winning. Re-use the WTA network to obtain the next point, directly access to all of the salient points. And then obtained a significant point of the basic points, using the region growing method for image segmentation, Eventually get all of the region of interest.Fourthly, In order to validate the effectiveness and practicability of our research of extraction method of region of interest, we develop an extraction system of region of interest based on matlab 7.0.Experimental results show that our method showed a higher effect than Itti method. Compare with the Itti method, our method was an increase of 4.2 percent in terms of accuracy. And was an increase of 28.8 percent in terms of time efficiency. It would be beneficial to improve the efficiency and accuracy of CBIR.