A Research On K-nearest Neighbor Classification-guided Iterative Regional Graph-cuts Segmentation Algorithm

Image segmentation,as a basic problem in image processing,which can divide an image into different regions,plays an important role in computer vision,object recognition,tracking,and image analysis.The graph cuts algorithm is an interactive image segmentation technique,which constructs energy functions based on region and boundary information and can achieve globally optimal results.Image object extraction from a complex background using the standard graph cuts algorithm often needs a number of user interaction information and give rise to segmentation errors as well.To fix this problem,a k-nearest neighbor(KNN)classification-guided iterative regional graph cuts segmentation algorithm was proposed in this study.Firstly,with the mean shift algorithm,the original image is pre-segmented into multiple homogeneous regions,as super-pixels.A weighted sub-graph is constructed based on the seed markers to label the adjacent unlabeled regions.Meanwhile,self-training KNN classifier is used to evaluate the segmentation label confidence during each iteration,and high confidence label super-pixels are selected as the seed markers to guide the next iterative segmentation.Compared with other image segmentation,the segmentation experiments on the same experimental image group show that the proposed method has efficient accuracy and robustness.