The Image Parallel Processing Based On The Cluster Computer System

The image parallel processing (IPP) is an integrated digital information processing technology. It is mainly based on the parallel computing environments and algorithms at low cost to achieve high-performance computing in virtue of current equipments and algorithms. IPP is an important and enduring technology in computer calculation to deal with the large amount of data. The main purpose of IPP is to realize the fast and real-time image processing. The IPP method utilized in this paper divide the images into smaller blocks(task granularity) according to the number of nodes in the cluster and the tasks, then the master node broadcasts the division information of the tasks to every slave node and the slave nodes execute the assigned task according to the requirements. Meanwhile, every slave node aslo sends the result data back to the master node and then it assembles the datum. Using this method, the image parallel processing has been simulated. The performance of the parallel processing simulation has been validated through the experimental data, and the merits of the image parallel processing are validated by the accelerating ratio and efficiency. The good greement between simulation and experiment suggests that this method is feasible. The research work and contents in this paper are briefly described as following:Firstly, the background, current status and significance of parallel processing in the IPP field have been summarized. The author has detailed the approaches about image enhancement, edging detection, featuring extraction, etc, introduced the parallel computing and parallel algorithms. The design process of master-slave model algorithm and the evaluation standards of measuring the parallel performance have been dicussed in detail.Secondly, the BSP computing model has been brought in and improved. Based on this improved BSP model the author has designed the system architecture of cluster to meets the requirements of real-time image parallel processing. At last, the image parallel processing based on the BSP computing model has been synthetically analysed. Through spatial domain processing methods such as image enhancement, edge detection, feature extraction, etc., the author has obtained the processing time of simulation using single and multi-computer parallel computing. Comparatively analysing the feasibility and effectiveness of parallel processing, also the author has simulated experiments through fast fourier transform and inverse transform of parallel processing in the frequency domain. The decomposition and reconstruction of experimental images has been done by using wavelet lifting algorithm. The accelerating ratio, efficency of the parallel computing in multiple nodes have been calculated and also their performance is been analyzed. It shows the method is real-time and valid, which is in good agreement with the experiment.