Research On Image Segmentation Algorithm Of Level Set Method Based On Saliency Regions

Image segmentation is a key from image processing to image analysis and very important in the research field of image. Image segmentation aims to divide image pixels with similar characteristics into the same region. It has been the focus of attention for a long time and has made thousands of different types of segmentation algorithm. However, there is no algorithm can be generally applied to all images. Further research is still needed.Level set method is an interactive segmentation algorithm. Through manual annotation, according to the energy minimization evolutions curve, that greatly improved the segmentation results. However, when facing the massive image data processing, human interaction will not be achieved. In order to obtain better segmentation information, people need debug repeatedly.Saliency detection algorithm is the abstraction of the human visual attention mechanism in the computer vision. Aiming at the hot topic of the saliency detection algorithm, this paper combines the salient region detection with level set method, proposing a new segmentation algorithm. Instead of manual interaction, the new algorithm uses the technology of automatic segmentation to obtain accurate segmentation results quickly. This paper mainly focuses on the following two aspects:First, the salient region image is obtained through saliency detection algorithm. SF saliency detection is a kind of detection algorithm with a better result and faster speed in recent years. But it only considers the color information of the uniqueness and distribution, when foreground and background have the similar color, it is easy to get the wrong results. According to this, this paper adopts the method of global rarity to solve this problem, which will obtain more salient regions effectively and apply to the segmentation algorithm conveniently.Then, this paper studied the classical model of level set method-CV model. In order to solve the problem of artificial participation and sensitive to initial contour, this paper adds the saliency detection into CV segmentation algorithm. First, we regard the saliency region as the initial contour of the CV model segmentation algorithm. Then facing the problems of the integration, we make the further improvement and achieve automatic segmentation. At the same time, the evolution curve starts from the edge of the object. That will reduce the interference of background information and increase the segmentation accuracy and improve the operation rate.Finally, this paper analyses the improved algorithm from two aspects of subjectivity and objectivity. The experimental results show the validity of this algorithm.