Image Segmentation Algorithm Based On Normalized Cut And Its Improvement

Image segmentation is a technology and process of extracting the target of interest by using some feature information of the image,and dividing the image into several special and unique regions.It is not only the key step of image processing to image analysis,but also the hotspot and difficulty in computer vision and digital image processing.In recent years,image segmentation algorithm based on graph theory has become a new research direction because of its good segmentation property.Among them,image segmentation algorithm based on normalized cut combines with the global and local features of the image,and it can maximize the total similarity within the regions as well as the total dissimilarity between the different regions,but the segmentation speed is very slow and easy to segment the similar size regions.Therefore,this paper makes a new image segmentation algorithm based on its shortage.The main contributions of this paper are as follows:(1)Image segmentation algorithm based on the normalized cut and the existing improvement about it was studied and analyzed deeply,and we found that the combination based on graph partitioning and normalized cut criteria can make up for each other's deficiencies,and accelerate the speed of image segmentation and can obtain better segmentation results.However,the algorithm also exists some deficiency,and therefore this paper puts forward the corresponding improvement method.(2)This paper defined a new weight function,strengthened the characterization ability of two pixels between the degrees of difference,and also accelerated the weight calculation speed,and reduced the required memory storage space.(3)The threshold function is improved from two aspects.On the one hand,the penalty intensity of the small region is enhanced by the penalty parameter,and the sensitivity of the parameters to the magnitude of the weight value is reduced.(4)The region adjacency graph was improved mainly by introducing controllable parameters derived to obtain each coarse segmentation area is representative of a graph is the number of vertices,overcome the some special circumstances of the poor segmentation results.Finally,we select the images of different sizes,types and contents from the BSDS500,and do experiment with the algorithms of this paper and other algorithms,and carry on the analysis evaluation of the system from three aspects of the segmentation results,running time and memory space.The results show that the algorithm can get good segmentation results quickly,and save the memory space of computer.