The Parallel Image Denoising Research Based On Fourth-order PDE

Massive processing data and complicated calculation is inevitable in image processing area. At present, the digital image processing speed is difficult to satisfy the requirements of real-time. The parallel computing is the most effective technology in terms of improving image processing speed. Image denoising is a basic problem in image processing. Fourth-order partial differential equation denoising method has a good noise reduction effect, but its massive calculation affects real-time. Image processing's parallelization design is in order to improve image processing speed and expand the scale of the image processing. Therefore, parallel image processing algorithm research is currently very important and valuable direction.This research is mainly based on the fourth-order partial differential equations parallel algorithm for image noise removal models. Firstly, an improved fourth-order partial differential equations model is proposed. The new model constructed a four directions lip operator to measure the image information by introducing four directional derivatives and lip mathematical model in the original one, which can effectively remove noise, preserve edge details and reduce the error to accord with human visual, and constructed a fidelity controlled function by using the noise visibility function based on the structure characteristic of human visual system, which can further preserve the edge details and avoid estimating noise level factitiously. Experimental results show that the improved method has superiority in the visual effect and objective quality, which can better remove noise and preserve edge details characteristic. Secondly, a parallel algorithm was designed for the new model based on the MPI to improve the speed and efficiency of the algorithm, which started from the partition strategy selection, date division, super step division and several other aspects by analyzing the parallelism of fourth-order PDE denoising algorithm, and then realized the simulation based on experimental environment of the built cluster. Experimental results show that the parallel algorithm not only improves the computing efficiency, shorts the running time and has high parallel performance, but also basically obtains the comparable image processing effects compared with the traditional serial denoising algorithm, which reflects the necessity and advantages of the parallel algorithm. Thirdly, a further optimization for the parallel algorithm was made, which adopted the hybrid parallel programming model to realize the multi-grain parallel by using the two parallel structures in inter-node and intra-node. In addition, the hybrid parallel time model of two-dimensional image was abstracted to analyze the performance bottlenecks of the parallel algorithms in order to search the optimal regularity and solve similar application problems. The simulation was implemented with the same experiment platform and the experimental results show that the optimized parallel algorithm is able to obtain the same image processing quality, but its parallel performance is better, which demonstrates the superiority of hybrid model parallel algorithm. Finally, the optimal parallel denoising algorithm was applied in satellite cloud image in the field of weather forecasting, the experimental results validate the practical value of the subject.