一种基于角点检测的图像密集匹配算法

Dense matching algorithm based on corner detection

提出了一种鲁棒的图像自动立体匹配算法。利用Sobel算子对图像中的像素点进行检测,若是边缘点,则使用最小同值分割吸收核方法判断该点是否为角点。在两幅待匹配的图像间计算角点的梯度大小、梯度方向及灰度等的相似度,去除无法对应的角点,建立起待匹配图像中角点的对应关系,并计算基础矩阵。对基础矩阵进行迭代,去除误配点,计算出较精确的基础矩阵。由对极几何约束,采用动态规划方法,寻找左右两幅图像在对应极线上的所有像素点之间的对应,从而建立起两幅图像间像素点的密集匹配对应关系。试验结果表明,算法效果满意。

A robust automatic dense matching algorithm is proposed. For every pixel of an image, the Sobel operator is used to determine whether the pixel is on the edge. If it is, the smallest univalue segment assimilating nucleus approach is used to decide whether the pixel is a comer. The comers that can not be matched are discarded across the left and right images, and the initial correspondence relationship by the retained comers is established between the corresponding comers of the left and right images according tothe similarity of the gradient magnitude, gradient orientation and gray. The fundamental matrix is computed. The initial matching result is iteratively improved by using the fundamental matrix to discard outliers, then the ultimate matching result is obtained and the fundamental matrix is accurately calculated. The matching relationship between the pixels on the corresponding epipolar lines of the left and right images is searched using dynamic programming approach in terms of epipolar geometry constraint. Therefore, the dense corresponding relationship of all the pixels between the left and the right images are established.