一种灵活的立体图像校正方法
A Flexible Rectification for Stereo Images

针对现有立体图像校正算法在校正图像时容易引起分辨率及图像内容损失的问题，提出了一种新的立体图像校正算法．首先突破了传统基于标定的校正算法必须使摄像机水平轴平行于光心连线方向的限制，提出了一种能够灵活选取坐标轴方向的方法．然后将光心位置差异引起的垂直视差分解为两个与光心位置相关的分量，并利用最小二乘法估计出一个相似变换和一个仿射变换，完成对摄像机光心的校正．实验结果表明，所提出的算法生成的左右立体图像的分辨率及内容损失大约近似仅有欧几里得方法的1/2，且对垂直视差的消除要优于欧几里得方法．
To solve problems that existing stereo rectification algorithms often induce a loss of resolution and distortions of image，a novel stereo image rectification algorithm was proposed in this paper．Firstly，this paper presented a more flexible method for the selection of the axis directions，breaking the limitation of the traditional calibrationbased rectification algorithms，which requires the horizontal axis of the camera to be parallel to the camera baseline． Then the vertical parallax induced by different optical center positions was decomposed into two components related to the optical center position，and the least squares method was used to estimate the according similarity transformation and affine transformation to finally complete the rectification of the camera optical center．The experimental results showed that the resolution loss and stereo distortions obtained by the proposed algorithm only reached approximately half of the Euclidean approach．In addition，the proposed algorithm outperformed the Euclidean approach in vertical parallax elimination