非标定图像对最小化畸变几何校正方法研究

图像校正需要对图像进行透视变换,不可避免会产生欠采样或过采样问题,同时射影变换会引起图像扭曲,造成图像对之间相似性丢失.为了降低图像校正后的畸变程度,重点研究了图像校正过程中不同类型图像畸变问题.通过图像对规范化简化基础矩阵,并通过考虑几何约束简化校正变换矩阵;然后求取射影变换权值变化测度之和的极值以实现最小化射影畸变;最后采用Jacobian行列式度量校正前后的局部区域变化,求得最小化重采样效果的矩阵系数.采用该校正变换矩阵对非标定图像对进行平面校正可以减轻射影扭曲程度和重采样效果.

Minimizing deformation geometric rectification of non-calibration stereovision

Projective transform in the process of image rectification produces under-sampling or over-sampling. To reduce deformation from image rectification, multi-type image deformation was studied emphatically. Firstly, a pair of images was normalized to simplify fundamental matrix, and the geometric constraint was applied to simplify rectification transform matrix. Secondly, the extremum of sum of projective transform weight change was solved to minimize projective distortion. Lastly, the rectification transform matrix was solved to minimize resampling effect which was weighed by Jacobian determinant. A planar rectification on epipolar geometry from the matrix could minimize the deformation including the projection distortion and the resampling effect.