在传统的多曝光图像融合方法中,一旦目标发生移动则会在最终融合图像中出现“鬼影”现象。现有的去“鬼影”算法大部分都继承了参考图像中的大量信息,倘若参考图像中出现曝光不足/曝光过度现象,便会影响到最终的融合结果。基于此,提出了一种基于图像块分解的多曝光图像融合去鬼影算法。首先将参考图像划分为曝光正常及曝光不足/过度两大区域,并有针对性地对这两部分区域进行处理。为了更加精准地检测出鬼影区域,将多曝光图像块分解成信号结构、信号强度和平均强度3个概念相独立的部分,采用图像块结构一致性检测的方式来进行鬼影检测。最后,去除结构不一致的图像块并对这3个部分分开融合,重构所需图像块并将其聚合至最终融合图像。实验结果表明,与现有的去“鬼影”算法相比,所提算法取得了更好的视觉效果,且计算效率得到了较大提升。

In traditional multi-exposure image fusion methods, once the target moves, the phenomenon of “ghosting” occurs in the final fused image. Most existing de-ghosting algorithms inherit substantial data from the reference image. Once the underexposure/overexposure occurs in the reference image, the final fusion result is affected. To remedy this, herein, a multi-exposure image fusion de-ghosting algorithm based on image block decomposition is proposed. First, the reference image is divided into two areas, i.e., normal exposure and underexposed/overexposed areas, both of which are individually processed. To detect the ghost area more accurately, the proposed algorithm decomposes the multi-exposure image block into three independent parts, i.e., signal structure, signal intensity, and average intensity. Ghost detection is then performed by detecting structurally consistent image parts, following which inconsistent parts are removed, the three image parts are fused, and the required image parts are reconstructed and added to the final fused image. Experimental results of this algorithm's validation show that compared to existing de-ghosting algorithms, the proposed algorithm achieves better visual effects and improves computational efficiency.