Design And Implementation Of Parallel Algorithms Image Segmentation For CUDA

As an important image technology in the field of computer vision, image segmentation is the basis of image analysis, recognition and understanding, which has got widespread attention both on the academic research and practical applications. However, most current image segmentation algorithms are serial algorithms, which lead to more iteration and lower efficiency. This paper mainly research on paralleling the image segmentation algorithm and improving the efficiency of image segmentation.In recent years, GPU has got great attention because of its powerful general-purpose parallel computing. It is widely used in the image processing research. Meanwhile, Compute Unified Device Architecture CUDA based on GPU is launched by NVDIA. With CUDA, C, C++code can be sent straight to GPU, hardware source of GPU can be directly accessed, and the parallel computation of video card can be fully harnessed. It provides a convenient platform and environment for developers to develop scalable parallel programs. And because of that, we consider implementing the image segmentation parallel algorithms.The main work of this paper can be concluded as follows.First, the basic research and correlated technique of this paper are introduced. A summary of current image segmentation algorithms are made, of which the advantages and disadvantages are analysis. The software architecture, hardware architecture and programming model of CUDA are introduced in detail.Next, in order to make up the deficiencies of serial algorithms, we present a CUDA-based image segmentation parallel algorithm, which is based on the region growing. It divides the original image into some pieces and processes the iterations in the pieces firstly, and then combines the results of all the pieces. In addition, considering the features and the advantages of CUDA, some optimization strategies are also brought out to improve the efficiency of the parallel algorithm.At last, the implementation of the image segmentation parallel algorithm on CUDA is introduced, including the process, the architecture and the modules of the program. Detailed experiments comparing the serial and parallel algorithms are made, and the experimental data and the results are presented and analyzed.The experiments prove the advantage of CUDA in the field of massively parallel computing.