Color Image Segmentation Based On Improved Watershed Algorithm And NCut Algorithm

Image segmentation is an important and basic part in image processing and computer vision. Image segmentation will have a direct impact on the subsequent image processing and image understanding Most of the existing image segmentation algorithm are based on gray image,but with the development of image processing technology,only using brightness information to describe the image is difficult to meet the requirements of segmentation. Color image, which contains more information than gray-scale images, can better feedback object features such as texture, shape information, so handling color image is more in line with expectations. With the study about watershed algorithm and NCut algorithm on gray scale image,this paper respectively improves both of them, and proposed a new algorithm to segment color image. The main research work and innovate ideas are as follows:1. Despite using watershed algorithm to segment image can get the enclosed edge with a single pixel width, after which it need not the operation of edge connection,and this algorithm has the very good response to weak edge about image, but the algorithm is sensitive to the noise so that it is too easily to occur over-segmentation phenomena. For this problem, this paper improve the traditional watershed algorithm with the method of tagging real minimum. For a color image, with the preliminary processing we get the gradient image, then using butterworth low-pass filter and multi-scale morphological operator for further processing to wipe the noise off the gradient image and obtained mark image. Finally, we apply watershed transform to segment a new gradient image which can be obtained through extending the minimum operating in the marked image.2. Traditional NCut algorithms are based on the undirected graph. As weighting matrix dimension of undirected graph is higher, solving the characteristic equation becomes very difficult. In addition, the algorithm is also sensitive to noise.According to the above-mentioned drawbacks of NCut algorithms, this paper puts forward the improved a new method for image segmentation which com-bines improved watershed algorithm and NCut algorithms based on region. First,the method applies the improved watershed algorithm to divide color image into several small areas, which retains the information of local edge andtexture, and then, in Luv space,combined the pixel information with location information of each areas after pre-segmentation to build undirected graph which based on region. The weight matrix dimension of improved algorithm is massive lower than traditional NCut algorithms. In addition, this paper proposes a new adaptive method to calculate the weight of undirected graph. Experiments show that the new segmentation method which combined improved watershed algorithm with the NCut algorithm based on region has some robustness to noise, and can achieve the high resolution color image segmentation. The weights of the adaptive can accurately reflect the position information and pixel information of each regions.