Research Of Random Walk And Its Application In Leaf Image Segmentation

Random Walk is an emerging intelligent information processing technology in recent years.It has been widely applied into image processing,especially with excellent performance in image segmentation.Therefore,this paper mainly discusses about the random walk technology.It focuses on the important improvement model of random walk and its application in leaf image processing.In this paper,first of all,the random walk technique is comprehensively analyzed and organized.Then,an improved method based on random walk algorithm for segmenting leaf image under complicated background is proposed.The main contributions are shown as follows:1.Six representative random walk models have been listed and explained in detail,such as Grady's Random Walk model(RW),Random Walk using Prior model(PRW),Generalized Random Walk model(GRW),Scale-Space Random Walk model(SSRW),(Constrained Random Walk(CRW)and Random Walk with Restart(RWR).The applications of random walk in eight aspects,like image segmentation,image fusion,hole-filling,2D-3D conversion,image rendering,image enhancement,image registration,and image retrieval,are introduced.2.We propose an improved method based on random walking algorithm for segmenting leaf image under complicated background.First,K-means is taken to find the centroids of the pre-defined seed pixels.A random walking model is then built based on prior pixels(the centroid of the pre-defined seed pixel).Then,in order to adapt the model to more complex backgrounds,the proposed method integrates the results of the standard random walk algorithm,which makes the method robust.At last,self-adaptive threshold is used in the binarization of probability map.The optimal results are output.The work also discusses the performance of RGB and Luv color space under the random walking framework.3.The proposed method is tested on ImageCLEF 2013 database.Experimental results show that the proposed method improves the performance of image segmentation in both accuracy and stability,and can produce better leaf image segmentation results.