Research Of Saliency Region Detection Of Images

With the rapid development of the Internet techniques and information processing techniques, digital media information is expanding rapidly with the geometric growth rate. It’s not realistic to deal with the information by human. People are more willing to use the computer to process and analyze the image and video data to obtain useful information quickly. This presents a higher challenge to computer vision technology. In recent years, in order to carry out the intelligent processing of media data, human visual attention mechanism has attracted wide attention from researchers at home and abroad. The visual attention mechanism is refers to the humanity in the face of complex scene, human still be able to find the salient regions most important quickly and accurately. Through the visual attention mechanism, human visual system can be selected to be the most important part of the scene. In the field of computer vision, researchers imitate the visual attention mechanism to propose image saliency detection technology. The goal of image saliency detection is to obtain a saliency map which can accord with the attention of the human eye by the computation of significant value. Therefore, it is an important direction of the computer vision research field to detect the salient region of the image by the visual attention mechanism and computer vision technology. It is of great significance to improve the efficiency of image processing.In this paper, we will study and research the existing salient region detection algorithms. We will point out the shortcomings of the existing methods and propose some improved methods. The main contents of this paper are as follows:The RC method is one of the outstanding saliency detection algorithms, but the RC method uses the Graph-based method to divide pixel blocks, which makes the pixel block shape irregular and non-compact. So the divided pixel blocks cannot fit the object edge well. Besides, since the regional saliency calculated by the RC method is directly assigned to the pixels in the region, it will introduce error caused by pre segmentation. Therefore, if we use better pre segmentation and improve distribution of regional saliency, it will improve the effect of saliency map.In view of the above shortcomings of the RC method, this paper proposes a salient region detection method based on SLIC super pixel segmentation. The saliency map of this method has more smooth and clear boundary compared with RC method. It will not only highlight the target area uniformly but also reduce the influence of background in complex image. In the precision-recall curve, the precision of this method is higher than RC method in the case of the smallest recall.In order to solve the issues of the existing methods, a salient region detection method which joint central constraint and space distribution are proposed. It will overcome the area dependence problem in traditional method. The generated maps not only have the characteristics of full resolution and clear boundary, but also can effectively restrain the influence of the complex background, and improve the performance of the detection method.