Parallel Implementation and Optimization of Haze Removal Using Dark Channel Prior Based on CUDA

Haze Removal Using Dark Channel Prior is one of dehazing methods with good effects, but its disadvantage of high time complexity limits the extent of its applications. In this paper, we present its parallel implementation and optimization based on the GPU, which greatly reduces the execution time. We firstly implement the basic parallel program, and then optimize it to obtain performance acceleration with more than 20 times. We introduce the method of "guide image filter" to Haze Removal Using Dark Channel Prior, instead of "soft matting" method, which largely reduces memory requirements and the computation complexity of the original algorithm. While paralleling and optimizing the algorithm, we improve some parts of the original serial program or basic parallel program according to the characteristics of several steps, and propose a novel method of selecting atmospheric light and a new parallel method of calculating accumulative sum with keeping intermediate results, which reduce the computation complexity of counterpart and increase the parallel degree.