Research On Image Segmentation Algorithm Combining Saliency And Level Set

Image segmentation is one of the most important techniques for detecting objects and analyzing images in the field of computer vision.However,due to the problems of rich color,intensity inhomogeneity and complex texture in real-world images,accurate image segmentation is still a huge challenge.The active contour model based on the level set method has become a mainstream image segmentation method because of its flexible topology changes and excellent segmentation performance.This paper conducts in-depth research on several key issues of existing level set methods,mainly including the following aspects:(1)Aiming at the problem that the existing level set method cannot segment images with severe intensity inhomogeneity and sensitive to initial contour.A weight function combining global and local information of the image is proposed,and the weight function is integrated into the level set energy function.The weighting function can weaken image intensity inhomogeneity and adjust region fitting information adaptively during contour evolution.Experimental results show that the proposed algorithm segmentation performance is better than other similar algorithms,which is not sensitive to initial contour.(2)Aiming at the problem that the existing level set method cannot accurately segment natural images with complex texture,a texture embedding level set method is proposed.Firstly,this method uses the image gradient information to calculate the local covariance matrix,and the image texture feature is extracted from the convariance matrix to construct level set texture term.Then,a higher-order edge operator is proposed to capture object boundary in complex natural images to construct the level set edge term.Finally,the three items of image intensity,texture and edge are integrated into the level set formulation to segment natural images,and the level set semi-implicit gradient descent method is derived to guide the evolution of contours more stably.Compared with the similar algorithms,the texture embedding level set method has better segmentation results for natural images.(3)In order to further improve the segmentation accuracy of complex natural images by the level set method,an energy function combining saliency and level set is proposed.Unlike most existing level set methods,this paper does not simply use the saliency map as the input image of the level set method to constitute saliency term.Instead,the saliency is used to obtain the background prior,so that the contour can quickly capture the object position and reduce the interference of complex background during contour evolution.This algorithm further improves the segmentation effect of level set method for natural images.