中值直方图均衡的动态场景多曝光图像融合算法

Multi-Exposure Image Fusion Algorithm for Dynamic Scene with Midway Histogram Equalization

为解决动态场景下多曝光融合图像出现鬼影的问题,提出一种新的动态多曝光图像融合算法。引入中值图像均衡对输入图像和参考图像的直方图进行处理,将获取的图像对做差分,并对差分图进行阈值分割和形态学优化得到运动权重图。中值直方图均衡可以为一对图像分配相同的直方图,同时保持其灰度动态,因此对多曝光图像对调整其亮度差异,有利于运动区域检测的准确性。通过强度映射函数将参考图像分别映射为各个输入图像的亮度,并将输入图像的运动区域替换为参考图像的一部分,得到具有亮度过渡自然的图像序列。在此基础上,对静态图像序列进行融合得到最终的融合图像。实验结果表明,该算法可有效地避免鬼影现象,且能够获得细节丰富、视觉效果良好的高动态范围图像,经该算法融合后的图像在标准差、边缘强度、相关系数和动态场景结构一致4个指标上与DGF、FMSD等算法相比具有明显的优势。

To solve the problem of ghosting in multi-exposure fusion images in dynamic scenes,this study proposes a dynamic multi-exposure image fusion algorithm with midway histogram equalization.First,Midway Image Equalization(MIE)is introduced to process the histograms of the input and reference images,the obtained image pairs are differentiated,and the difference image is subjected to threshold segmentation and morphological optimization to obtain a motion weight map.Midway histogram equalization can assign the same histogram to a pair of images while maintaining its gray dynamics as much as possible.Therefore,the brightness difference of a multi-exposure image pair can be adjusted,which is conducive to the accuracy of motion area detection.Second,an Intensity Mapping Function(IMF)is introduced to map the reference image to the brightness of each input image,and the motion area of the input image is replaced with a part of the reference image to obtain an image sequence with a natural brightness transition.Finally,the static image sequence is fused to obtain the final fused image.Experiments demonstrate that the algorithm can effectively avoid ghosting and obtain high-dynamic range images with rich details and good visual effects.In addition,images fused using this method have obvious advantages over the DGF and FMSD algorithms concerning standard deviation,edge strength,correlation coefficient,and dynamic scene structure.