The Research Of Cow Image Segmentation Based On Bidirectional Matching Of SIFT And Region Merging

Image segmentation is a typical problem in computer vision.Our research is mainly used on cow recognition system for video processing.,in which the effective segmentation of the cow's image is the basis of the system.At present,the difficult problem of image segmentation is how to distinguish the foreground and background of image automatically.The interactive image segmentation needs to mark the label about object and background on the image manually,which requires higher operator and the segmentation result is unstable.In addition,the region-based image segmentation often exist over-segmentation,making the segmentation results error.This paper mainly studies and discusses the above problems of image segmentation,and proposes a cow image segmentation algorithm based on bidirectional matching of SIFT and region merging.The main innovations of this paper are reflected in the following aspects:(1)In the interactive image segmentation process,it is usually necessary to manually mark each image to be segmented,and easily lead to instability of the segmentation result in complicated operation.In this paper,we present an algorithm that using the sample to guide the target area mark.Firstly,the binary cows image is taken as a simple sample to guide the extraction of the SIFT feature points in the target area of the cows and to match the feature points of the cows to be segmented.Therefore,the foreground area of the cows to be segmented is marked and achieves the automatic marking of the target area.(2)In order to improve the accuracy of matching and ensure the correctness of the marking results,this paper proposes a novel image matching algorithm.The proposed algorithm,combined the bidirectional matching of SIFT and RANSAC,using the idea of intersection for the traditional SIFT feature extraction algorithm and combined with the principle of random sampling consistency,which effectively improved the matching accuracy.(3)Aiming at the problem that only the foreground or the background region be marked can't be segmented in the image,a novel adaptive seed point selection method is proposed in this paper.The matched feature points were using as the seed points of marked areas,and the seed points of the foreground and the background were differentiated using concave hull by statistical methods.In addition,the pixel distributions marked by the traditional manual marking method are relatively concentrated,and the merge areas involved are relatively small,resulting in lower efficiency of merging.Therefore,in this paper,the distribution characteristics of SIFT feature are used to improve the execution efficiency of the algorithm greatly by expanding the number and scope of the merged regions,and completed the preliminary segmentation of the image.(4)For the image segmentation method often exists over segmentation phenomenon in the complex background,and this phenomenon is reflected in the cow's dorsal segmentation particularly.Therefore,in order to further improve the accuracy of cow image segmentation,this paper presents a back-optimized algorithm for cow segmentation based on the unique characteristics of over-segmentation regions.The algorithm first uses the algorithm for identifying the convexo-cancave of peripherals based on freeman chain code difference to extract the corner points from the cow's back contours,and then calculates the slope between adjacent inflection points.By comparing with a certain set threshold value,the unnecessary bulge point is removed,with iteration,so as to achieve the optimal segmentation of cow images.Finally,through the experimental verification of the data set of the cow image,this paper proposed the cow image segmentation algorithm can effectively solve the problems related to the cow image segmentation,and has good practical application value.