Image Segmentation Algorithm Using Shape Prior And Its Application

Image segmentation is a hot topic in the research of computer vision,pattern recognition and digital image processing.Image segmentation can be divided into two categories:interactive segmentation and automatic segmentation.In recent years,the interactive segmentation has made great progress,but existing complex operation,time consuming,laborious and sensitive to the interaction area and other issues when dealing with a large number of pictures,so the automatic segmentation method has been attended extensively by the researchers.In this thesis,an automatic image segmentation algorithm is designed.The key points of the foreground are automatically detected by the salient detection algorithm and the local shape matching algorithm,and they are applied to the star prior segmentation algorithm to realize the automatic image segmentation.The main contents and innovations of this thesis include the following parts:(1)The extraction of foreground key points plays a decisive role in the automatic segmentation of images.In order to automatically mark the foreground of the image,a foreground key extraction method based on local shape matching is proposed.The method uses the local shape matching to automatically mark the foreground and combines with the foreground marking method based on the salient algorithm to detect more foreground key points and is conducive to the subsequent automatic segmentation.(2)The combination of shape prior knowledge can improve the accuracy of image segmentation.Star shape is a universal prior knowledge,applying to many objects in nature.This thesis realizes Graph Cuts segmentation algorithm combined with the star shape prior,which effectively improves the segmentation quality.(3)The proposed image segmentation algorithm is applied to image special effects processing,this thesis presented an example-based silhouette stylization approach for images.Based on the image segmentation,the algorithm adopts different color processing strategies for foreground and background,so that the generated image has a similar style to the sample image.The experimental results demonstrate the effectiveness of the method.