Image Segmentation Based On The Nearest Neighbor Graph And Fitting Method

Image segmentation is to segment an image into mutually exclusive and meaningful regions.It is a basic problem of computer vision.In the past few decades,there have been a large number of publications on image segmentation,which are widely used in medical image processing,target tracking,recognition,image reconstruction and other fields.In the past image segmentation technology,many successful methods benefited from mapping image elements to the graph.Then use the relevant theoretical knowledge of graph theory to solve the segmentation problem in the discrete space.This paper mainly studies some characteristics and applications of image segmentation based on graph theory.The main work is as follows:(1)Aiming at the over-segmentation problem in the existing watershed algorithms,this paper proposes a watershed image segmentation method based on multi-scale morphological gradient reconstruction and the nearest neighbor graph merging criterion.Firstly,multi-scale morphological gradient reconstruction based on marker control is used for image pre-processing to eliminate redundant regional extremum and noise.Secondly,the superpixel regions generated by watershed transformation are merged by constructing the nearest neighbor graph merging criterion,so it improves the description ability of target features and enables the algorithm to obtain the feature information of background targets while dividing foreground targets.Finally,the experimental results show that the proposed method can combine similar regions better,and effectively solve the over-segmentation problem.Compared with the traditional watershed segmentation methods,this proposed method effectively weaken the over-segmentation problem and greatly improves the segmentation accuracy.(2)Research on color metrics suggests to use weighted Euclidean distance in the RGB color space to calculate the edge weights in the graph.It is one of the key parameters in graph construction and determines the segmentation effect.However,if you consider color images,it is not enough to consider the distance between two points.Some weights need to be associated with each component in the RGB color space.Therefore,this paper uses the weighted Euclidean distance to calculate the weight of the edge.In addition,based on the analysis and experimentation of the previous research,it is found that there is a certain relationship between the parameter k value and the difference between the maximum weight and the minimum weight of the image.Firstly,some images are simulated to obtain the relevant k value and weight data.Then,the data are fitted by the method of fitting,and a quantitative formula for estimating k value is obtained.Finally,we use the formula to experiment the image,and get a good segmentation effect.