Realization And Optimization Of Image Filtering Algorithm For Domestic Accelerators

Image filtering algorithm can eliminate or suppress the noise in the image without affecting the detail features of the image as far as possible,which plays a key role in the field of image processing.FT-M7002,as a high performance DSP with vector multi-core architecture,is suitable for the field of image processing.However,there are few researches on the development and optimization of image processing algorithm library for this processor,and the corresponding function algorithm library is not supported.In this paper,the research and implementation of image filtering algorithm for FT-M7002 processor is carried out,and the parallel optimization of the algorithm is carried out in combination with its architecture characteristics,which can fill in the blank of the application of image filtering algorithm on FT-M7002 platform,and it is of great significance to promote the adaptation of domestic hardware and software and expand the application of domestic high-performance DSP.The main work of this paper is as follows:(1)The overall implementation process of image filtering algorithm for FT-M7002 platform is designed.Through the in-depth study of FT-M7002 architecture,hardware characteristics and image filtering algorithm,the feasibility,difficulty and performance bottleneck of Gaussian filtering and median filtering algorithm on this platform are analyzed and elaborated,and the specific implementation process of the algorithm is given.(2)The parallel image filtering algorithm interface supporting multiple data types(8U\16U\32F)is implemented based on the characteristics of platform hardware and image filtering algorithm.To make full use of the hardware platform of the computing and storage resources,after analyzing the finder hot part,combined with the feature of image filtering algorithm program,through the vectorization rewrite,DMA transfer optimization,software pipelining,loop optimization methods improve execution efficiency of the core code and data to fetch efficiency,give full play to the computing power of the FT-M7002 platform.(3)A vector alignment interface suitable for FT-M7002 platform is designed.Since the vector comparison instruction of FT platform cannot directly obtain the numerical results required by the median filtering algorithm,the data type conversion instruction and multiplication instruction are combined to directly solve the numerical values required by the calculation process,and encapsulated into a general vector comparison interface,which further improves the algorithm performance.This interface mode has certain expansibility on FT-M7002 platform.In this paper,the correctness and performance of the optimized image filtering algorithm are tested.The results show that the optimized image filtering algorithm can achieve correct results for different input data on the FT-M7002 platform.The acceleration ratio of the core function of the Gaussian filter optimization algorithm is7.4-41.6 times,and the overall acceleration ratio of the algorithm is 1.39-2.75 times.The overall acceleration ratio of the median filter optimization algorithm is3.8-16.6 times.