Optimization And Parallelization Implementation Of Digital Image Segmentation Algorithm

With the rapid development of computer science and technology,multimedia technology and digital image processing technology is becoming more and more mature,the interaction of people and information become more frequently..Compared to text and speech,images can show more information,which has become an important information source of people.Image analysis is a method of observing region of interest to an image and is often used to create an overall description of an image.Image segmentation is an important part of image analysis,which has a great effect on understanding and analyzing image information.Only by using image segmentation to segment the interested area,further target can be done with the object,such as feature extraction and measurement operations.However,with the increase of the image resolution,the complexity of the graph increases,and the computation speed becomes more and more slowly,which has become an urgent problem to be solved.This thesis analyzes the theory and research status of image segmentation algorithm based on graph theory,and studies the basic knowledge and concrete realization of parallelization,and gives the optimization of the two steps.Firstly,aiming at the problem of large complexity of undirected graphs,a new energy function coefficient is proposed to evaluate the probability that the pixel belongs to the foreground and the probability that the pixel belongs to the background.On this basis,the number of t-links(an edge in an undirected graph connected to a source point and sink)reduced,discard one edge which has smaller weight in each pair of t-links,the other edge is weighted by the new energy function coefficient,reduced the complexity of undirected graphs without affecting the image segmentation results.Then,according to the problem of slow running speed of the algorithm,this thesis improves the specific process of push and relabel,divides the push operation into Push and Pull two steps independently,and introduces new relabeling operation to reduce the number of iterations and improve the algorithm running rate.Finally,the above two steps are combined to improve the overall efficiency of the algorithm.In this thesis,three benchmark images are used as test data,using the theory of parallelization,making full use of the characteristics of CUDA architecture to implement the algorithm in parallel and the comparison experiment is designed to verify the effectiveness of the improved algorithm.Finally,the experimental results show that the proposed algorithm can improve the efficiency of image segmentation algorithm.