高动态范围视频的多曝光图像序列快速融合  

Fast multi-exposure image fusion for HDR video

作　　者：朴永杰[1] 徐伟[1] 王绍举[1] 陶淑苹[1] 

机构地区：[1]中国科学院长春光学精密机械与物理研究所小卫星技术国家地方联合工程研究中心,吉林长春130033

出　　处：《液晶与显示》2014年第6期1032-1041,共10页Chinese Journal of Liquid Crystals and Displays

基　　金：国家863计划项目(No.2012AA121502)

摘　　要：使用多次曝光法进行高动态范围视频获取时,需要将多幅低动态范围(LDR)图像进行快速融合,并满足视频融合系统的实时性要求。本文提出了一种改进的快速多分辨率塔形分解融合算法,在不降低原有算法融合质量的前提下极大减少计算量。算法分别从金字塔分解卷积核、高斯系数扩展插值方法、拉普拉斯系数融合规则、高斯系数融合规则等方面进行改进,简化了塔形分解、融合和重构过程。该方法应用在某运载火箭视频采集系统中,实验结果表明,该改进算法稳定性强,融合质量高,并极大地减少了计算量,运算速度提高了10倍以上,在TMS320DM642DSP上一帧512×512图像的融合时间约为8.9ms,满足25fps高动态范围视频融合的实时性要求。Fast fusion of multiple low dynamic range（LDR）images is essential for the real-time performance of a high dynamic range（HDR）video acquisition system using multi-exposure method.This paper proposed an improved fusion algorithm based on multi-resolution pyramid decompose.The algorithm dramatically reduces computing complexity while preserving the fusion quality of the original method.The proposed method improves the real-time performance in the following aspects：pyramid decompose convolution core,Gaussian coefficients expansion method,Laplacian coefficients fusion rule,Gaussian coefficients fusion rule.The computing complexity of pyramid decomposes,fusion,and reconstruction is reduced.Experiment results shows that the proposed method is stable and got good fusion results.The method is applied to a launch vehicle observation system.The computing speed is boosted 10 times and it takes 8.9ms to fuse one 512×512frame on TM320MD642 DSP,fulfilling the real-time requirement of 25 fps HDR video acquisition system.

关 键 词：高动态范围视频 HDR 多曝光 图像融合 塔形分解 

分 类 号：TH703[机械工程—仪器科学与技术] TP391[机械工程—精密仪器及机械]