The study on infrared image parallel computing based on improved BSP model

This paper analyzes the characteristics of parallel algorithm for Image and real-time processing of image which base on the cluster-computer. An improved double-sync(the bulk synchronous and the cycle of synchronization) BSP model is proposed, which provides the capability of the real-time processing. Under the clustered computer system, we adopt the leaping task allocation algorithm and the method of border extrapolation and interpolation with which the problem of loss of super-step in real-time parallel computing is solved. Moreover, the paper sets up the cluster computer system architecture for real-time image processing, and the parallel algorithms of image by application of lifting wavelet transform has been implemented. The same time, we build the speed up-ratio model and then analyzes the performance of the calculation. In addition, the improved BSP model, the leaping task allocation mechanism and the border task method of the validity have been verified by experiments, which are effectual for increasing the real-time of image.