Optimization And Application Of Image Restoration Algorithm On BWDSP

Image restoration is to use the prior knowledge of the degradation process to restore the image before it is degraded.At present,a lot of research has been conducted on accelerated image restoration algorithms,and it is found that implementing image restoration based on high-performance processors is also an effective acceleration method.There are many researches on foreign processors,but few researches on accelerated image restoration algorithms based on domestic DSP.Therefore,this paper proposes a non-blind deconvolution image restoration algorithm based on the domestic high-performance processor Bowei DSP 1042(BWDSP1042).By optimizing key algorithms that take a long time to transplant on BWDSP1042,it can ensure the quality of the restored image at the same time,to minimize the overall running time of the algorithm.At the same time,to a certain extent,it is of practical significance to break through the monopoly of foreign high-end core technologies over my country.This paper optimizes and improves the three key algorithms of image feature point extraction,FFT algorithm and Wiener filtering,which are time-consuming in image restoration algorithms.Among them,improve the exponential function in the image feature point extraction at the algorithm level,and the results show that the operating efficiency of the traditional algorithm is increased by 3.5 times.In addition,the FFT algorithm and the matrix multiplication operation in the Wiener filter are based on the BWDSP1042's SIMD and VLIW hybrid architecture,instruction layout characteristics,and application function library software development and design optimization technology for on-chip optimization.In order to fully tap the underlying resources of BWDSP1042,use its computing components and instruction layout to the greatest extent,this paper is based on BWDSP1042 transplanting the key algorithm that takes a long time in image restoration,and verifies the function of the transplantation algorithm on the test platform independently developed by our country.The simulation results show that when the relative error of the exponential function program,FFT algorithm and matrix multiplication operation results are controlled within the 10-7 order of magnitude,under the same excitation,compared with the optimization based on BWDSP1042 before transplantation,the operating efficiency of image feature point extraction,FFT algorithm and Wiener filtering algorithm are about 4.3 times,8.1 times and 8 times higher respectively,which proves that the running time of the key algorithm that takes a long time is significantly reduced after optimization and transplantation.In this paper,a series of optimized and transplanted image restoration and efficient algorithms are applied to motion blurred images,and the image restoration algorithm based on BWDSP1042 is analyzed and evaluated from the two indicators of the restored image quality and the running time of the algorithm.The experimental results show that compared with blurred images,the image quality evaluation indicators MSE,PSNR,and SSIM are increased by approximately 5 times,1.3 times,and 1.6 times,respectively.By optimizing the key core algorithm that takes a long time to transplant,the overall running time of the image restoration algorithm is approximately 4 times shorter than the theoretical running time,and the actual running time of the original algorithm is shortened by 2.2 times.The simulation result proves the necessity and efficiency of optimized transplantation of the image restoration algorithm on the domestic processor BWDSP1042.Under the premise of ensuring the clarity of image restoration,the operating efficiency of the overall image restoration algorithm is significantly improved,and at the same time,it is helpful to promote the further commercialization of BWDSP1042 has practical significance.