Image inpainting using anisotropic heat transfer model based on CUDA
Zhou Bing-yua · Jisuanji yingyong yanjiu · 2014
This paper proposed the GPU parallel repair algorithm based on irregular image structure and texture for lack of texture information in digital image restoration processing of cultural relics and the problem of high computational complexity.On the basis of existing research,it modified the heat transfer partial differential equation model and highlighted considering irregular image structure and texture information to the influence of heat transfer in the model design,by increasing the important image recovery information to improve the effect of the repairment.However,a large number of irregular texture information processing reduced the efficiency of image inpainting.So in order to ensure real-time repair effect of the algorithm,through the restore process of GPU parallel Jacobi iteration process to improve the efficiency of the data intensive computing which ensured that the performance of image inpainting.Finally experiments show that the image inpainting model improves the effect of repairment and adds the universality of the application.Compared with serial algorithm at the same time,the calculation speed increases by more than 20 times.The parallel image inpainting method has good application value.