Saliency Detection Based On Location Prior And Superpixels

With the rapid development of internet and multimedia technologies, the amount of digital multimedia is growing exponentially. The massive multimedia data bring great convenience for people’s entertainment, education and commerce. However, they also make some new challenge for the existing multimedia processing technology. Saliency detection methods can automatically predict, locate, mine the important visual information, thus it can help computer effectively select important information from the massive visual data. Considering the computer vision system’s demands, we make research on the key technology of saliency detection. The main contributions of this paper are described as follows.First, the existing methods which detect the position of visual fixation point only analyze the local scarcity in the images, the background interference can be mistaken as significant prospects when the contrast is large. To solve this problem, this paper presents a new method of saliency detection based on Gestalt theory. We compute saliency using the characteristic that the salient object is always surrounded. Its essence is to analyze the global topology of the image’s foreground and background, which can effectively suppress the influence of the local scarce regions in the background. This methods’detection of visual fixation point position can be merged with other methods as a kind of location priori. The traditional methods of local priori usually just highlight the center area, and would fail if the salient object is far from the center. Our methods can overcome this problem.Second, some detection methods are based on global contrast in pixel level. These methods’ computational complexity is high, and they cannot explicitly express the salient object’s contour when the color is confusion and the texture is complicated in the image. To overcome these problems, we first propose a novel superpixel clustering method, SSLC. We make proper processing on the image, and compute the saliency value based on combing the super pixels’ uniqueness and spatial distribution’s influence. Then we propose a more reasonable framework. We merge the local priori map which is producing by the method based on Gestalt theory with the saliency map which is producing by the method based on computing superpixels’global contrast in order to obtain the more optimal saliency map.