两种新的光照方向估计方法  

Two New Approaches for Illuminant Direction Estimation

作　　者：杨杰[1] 邓志鹏[1] 郭英凯[1] 李介谷[1] 

机构地区：[1]上海交通大学图象处理与模式识别研究所,上海200030

出　　处：《上海交通大学学报》2002年第6期894-896,共3页Journal of Shanghai Jiaotong University

基　　金：国家自然科学基金资助项目 (3 0 170 2 74)

摘　　要：光照方向估计是计算机视觉的重要研究内容 .提出了两种新的光照方向估计方法 :改进的Pentland方法和加权平均法 .前者修正了 Pentland方法在数学上的缺陷 ,利用了图像中不同区域对光照方向估计的贡献 ,其性能比原方法有较大改进 ;后者利用图像中奇点的作用 ,并且考虑了图像中不同地形区域的特征 ,是一种稳定精确的光照估计方法 .Illuminant direction estimation is an important research content of computer vision. The paper analysed the current approaches for illuminant direction estimation, presented two improved approaches for illuminant direction estimation: improved Pentland's approach and approach of weighted mean. The former has overcome some drawbacks of Pentland's approach in mathematics. It makes full use of the different areas in an image which contribute illuminant direction estimation and improves the performance of the original approach. The latter makes full use of the effect of singular points in an image, and considers the characteristics of different landforms in the image, so it is a stable and accurate approach for illuminant direction estimation. The effect of those two new approaches was proved by experiments.

关 键 词：计算机视觉 光照方向估计 图像处理 Pentland方法 加权平衡法 图像微分 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]