Parallel Image Processing Based on CUDA

CUDA (compute unified device architecture) is a novel technology of general-purpose computing on the GPU, which makes users develop general GPU (graphics processing unit) programs easily. This paper analyzes the distinct features of CUDA GPU, summarizes the general program mode of CUDA. Furthermore, we implement several classical image processing algorithms by CUDA, such as histogram equalization, removing clouds, edge detection and DCT encode and decode etc., especially introduce the first two algorithms. If we donpsilat take the data transfer time in experiment between host memory and device memory into account, as the image size increase, histogram computation can get a more than 40x speedup, removing clouds can get an about 79x speedup, DCT can gain around 8x and edge detection more than 200x.