Abstract: Image sequences are exposed differently in the same scene, often in areas that are underexposed or overexposed, thereby causing loss to the highlight details or the shadows. In order to solve this problem, multi-exposure HDR images reconstruction based on multi-scale detail fusion is proposed, which considers three measure factor of image, such as contrast, saturation and well-exposedness. This thesis maps the decomposed weight Gaussian pyramid through Dirichlet function, and assigns maximum weights for areas with rich information, then the HDR image can be reconstructed by the Laplace pyramid, which contains the maximal detail information and minimal distortion.

Key words: high dynamic range image, multi-exposure image, image fusion, multi-scale

摘要： 同一场景不同曝光的图像序列，常出现曝光不足或曝光过度的区域，造成高亮或阴暗处的细节损失。针对这一问题，提出的多尺度细节融合的多曝光高动态图像重建方法，根据图像的对比度、饱和度、适度曝光量等三个测度因子生成原始多曝光图像的权重图，对分解的权重高斯金字塔进行Dirichlet函数映射，保证信息丰富区域权值最大，通过拉普拉斯金字塔重建，使得融合图像所包含的细节信息最大化并且最大限度地减少失真。

关键词: 高动态范围图像, 多曝光图像, 图像融合, 多尺度细节