对视差估计中摄像机增益差的校正(英文)

Correcting for the Gain Difference of the Cameras in the Intensity-Based Parallax Estimation

本文讨论一种基于亮度对立体图对左右图象的视差进行估计的方法。因为最简单的基于亮度的方法 ,如块匹配法 ,在尺寸块较小时对噪声比较敏感 ,所以近来人们提出了一种基于非线性扩散的比传统块匹配法有更好的噪声抑制性能的改进块匹配法。本文将探讨这种新方法在由左右摄像机增益差而导致立体图对亮度和反差不同时的情况。首先我们指出改进块匹配法与传统方法相比对上述亮度和反差方面的差别更不敏感。其次我们介绍一种基于立体图对统计特性的称为增益控制的方法以改进视差估计算法的性能。

In this paper an intensity based meth od to estimate the parallax between the left and the right image of a stereo pair is studied. Since the simplest intensity based method, i.e. block matching, is sensitive to noise, when small block sizes are required (as e.g. for computer vision purposes), a new method, improved block matching, based on the non linear diffusion process, was recently developed. It was shown that this new method has better noise suppressing properties than traditional block matching. Here, it is investigated how the new method reacts to a difference in average luminance and contrast of the left and the right image, caused by a gain difference in the left and the right camera. First, it is shown that improved block matching is less sensitive to these differences than traditional block matching. Secondly, a correcting mechanism, called gain control, based on the statistics of the left and the right image of the stereo pair, is shown to improve the performance of the parallax estimating algorithm.