Study And Implementation Of GPU Acceleration On Neighborhood Filtering Algorithm

The image in the process of imaging polluted by noise,which already affect the quality of the images.Background of this paper arising from ice cream version of visual inspection system to collect the images were polluted by noise,which make the detection system after testing steps can not be enforced,so it need to filter to the acquisition.This article using the method of filtering is Neighborhood filter.Neighborhood filter are divided into two kinds,one is Linear neighborhood filter,another is Non-linear filter.Because the filtering algorithm to the whole image,lead to the algorithm execution efficiency is low.Introducing concept of GPU acceleration image neighborhood filtering algorithm,its aim is to effectively accelerate the speed of execution of neighborhood filtering algorithm,in order to the algorithm to the efficient implementation of concise.Those are GPU acceleration algorithm of Box filter、Mean filter and Gaussian filter. Compared with execution efficiency with three algorithm of Linear neighborhood filters are run by Open CV CPU, the performance of GPU is efficient and speedy, especially for basic calcuation template convolution and others which need much more cycle processing.It is not like Linear filter,Non-linear filter has good preserving edge and removing noise, and it improves the resolution of the image, but Open CV GPU module involves little algorithm on the size of Non-linear filter, so that Non-linear filter is rarely applied in the filed of GPU.In order to adjust to GPU acceleration to Non-linear filter, the paper designed two programs to Bilateral filter algorithm of Non-linear filter in GPU acceleration, one way is the program of speeding up based on CUDA, another way is the program of speeding up based on combination programming. On the basis of the implementation of the algorithm, comparing their similarities and differences, analyzing their own characteristics and the key points of implementation process. And then give a better Bilateral filter algorithm of GPU acceleration.Through the research to Bilateral filter algorithm of GPU acceleration, we can know the above two programs on image processing effect and the execution efficiency had no difference, but in size of the structure of the programming, the combination programming program of Open CV + CUDA is easier and speedier to the programming on CUDA platform simply, which can complete GPU acceleration of Bilateral filter algorithm efficiently.Finally,the paper would combine programming of Open CV GPU + CUDA effectively applied to ice cream board of visual inspection system,effectively improve the speed of detecton system.And it reflected the significance of the research to the image processing parallel acceleration fully, so the research has practical value. Because of versatility of the experiment,it can be widely used in other areas related to image processing, which can promote the development of more other industries.