Visual Saliency Detection Based On Context And Background

It is well known to all the people that animal vision systems can effortless and efficiently distinguish salient regions from a cluttered background. However, it is still of great interest to obtain a salient region efficiently and robustly in computer vision system. A salient region in an image is the part which can express the information of the image best and can also catch most of our attention.In this paper, we propose a saliency detection algorithm based on combination of context and background information. Different from other methods which use context information or background information only, we select a proper joint point which can make both methods contribute positive energy to the final saliency detection method. Firstly, we use two methods to segment image into several blocks when preprocessing images. One is irregular block, the size between blocks has a very large gap, but the boundary of the image block and boundary of salient object in image is similar. The other is regular block, the size of blocks and color and spatial information is similar. On the purpose of winding speed, we replace the information of a block with the information of the center pixel of this block.Secondly, we introduce one context based method and two background based methods. Thirdly, we proposed a high-pass filter which is suppression to the background and can reserve foreground. Better combination result can be obtained based on this filter. Finally, we improve the convex hull in Bayesian framework to locate the object more accurately.We make a comparison between our method and several mainstream approaches on an international public dataset. The result shows that our method can achieve preferably performance.