基于图像内容自适应矩阵完成的高动态范围成像方法

发布时间：2018-05-20 15:13

本文选题：高动态范围成像 + 运动先验知识　；参考：《光电工程》2017年11期

【摘要】：高动态范围图像合成中容易出现鬼影现象,而传统的矩阵完成算法没有考虑多曝光图像序列的运动特征,导致高动态范围图像中的鬼影未能完全去除。针对这一问题,本文提出一种基于低动态范围图像内容自适应矩阵完成算法以去除高动态范围图像中的鬼影。首先,根据图像的亮度和色度信息,确定低动态范围图像的运动区域;然后根据这一运动先验信息,调整矩阵完成算法过程中的正则化约束强度,从而得到每幅低动态范围图像的背景信息;最后,考虑到不同曝光度的图像在每一区域所包含的细节不同,采用与曝光相关的融合策略合成去鬼影的高动态范围图像。实验分别采用简单背景图像序列和复杂背景图像序列,结果表明,所提出的算法相比于奇异值部分和最小化矩阵完成算法,能取得更好的合成效果,适用于复杂背景下的高动态范围图像合成。
[Abstract]:Ghosts are easy to appear in high dynamic range image synthesis, but the traditional matrix completion algorithm does not take into account the motion characteristics of multi-exposure image sequence, resulting in the ghost in high dynamic range image can not be completely removed. To solve this problem, this paper proposes a content adaptive matrix completion algorithm based on low dynamic range images to remove ghosts from high dynamic range images. Firstly, according to the luminance and chroma information of the image, the moving region of the image with low dynamic range is determined, and then the regularization constraint intensity in the algorithm is adjusted according to the prior information of the motion. The background information of each low dynamic range image is obtained. Finally, considering the different details contained in each region of the image with different exposure, a fusion strategy related to exposure is used to synthesize the high dynamic range image. The experiment uses simple background image sequence and complex background image sequence respectively. The results show that the proposed algorithm is more effective than the singular value partial and minimized matrix completion algorithm. It is suitable for high dynamic range image synthesis in complex background.
【作者单位】：宁波大学信息科学与工程学院;南京大学计算机软件新技术国家重点实验室;
【基金】：国家自然科学基金资助项目(61671258) 浙江省自然科学基金资助项目(LY15F010005,Y16F010010)
【分类号】：TP391.41

【相似文献】