Parallel implementation of low light level image enhancement using CUDA

Enhancement algorithms can make low light level images have a clear visual effect like the one captured during the daytime, but due to high complexity and generous computational cost, low light level image enhancement algorithms are usually difficult to meet real-time requirements which make it difficult to be widely used in practical application. For this situation, a parallel optimization algorithm of low light level image enhancement using CUDA is proposed. Enhancement algorithm based on de-hazing technique is used and on CPU-GPU heterogeneous platform the part of atmospheric light estimation which is not suitable for parallel computing is improved to obtain high parallelism degree. By comparing the performance of the algorithm on GPU with CPU, we indicate that the algorithm proposed has a significant improvement in execution speed while maintaining the visual effect of the traditional algorithm.