Graphic Processor Unit Acceleration of Multi-exposure Image Fusion with Median Filter

摘要

A fast Graphic Processor Unit(GPU)accelerate algorithm of multiexposure image fusion with median filter is presented in this paper.The proposed algorithm fuses images in YUV space instead of RGB space compared to traditional image fusion method.Furthermore,in YUV space the brightness components and the chromatism components were weighted fused separately with median filter.At last the filtered images were transferred to RGB and merged to the final fusion image.In the GPU acceleration part,three parallel methods were proposed,including sequence images concurrent execution,adjacent kernels merge,and parallel median filter techniques,to expand the concurrency of the algorithm on the GPU platform.In the experimental results,a 16–21 times speedup was obtained compared to the CPU implementation and up to 60 fps performance was achieved in a 1000*1000*6 multi-exposure sequence image fusion case.The results in the experiment demonstrate the high efficiency and high availability of our proposed method.