基于角点匹配的鲁棒图像镶嵌方法

为了获取场景的宽视场表示,提出了一种鲁棒的图像镶嵌方法.该方法对图像序列中的各个图像进行角点提取,再利用归一化相关方法进行角点的初始匹配.由于初始匹配中包含有大量的出格点,直接采用最小二乘法得不到正确的图像平面间的变换模型参数,为此,采用了鲁棒随机采样算法来估计图像平面间的变换模型参数.该算法能够有效地剔除初始匹配中的出格点,获得精确的匹配点子集.然后利用这些精确匹配的点集来计算变换模型参数,从而实现了图像的正确镶嵌.整个过程无需人工干预,均由计算机自动完成.对真实图像的试验结果表明了该方法的有效性.

Robust Image Mosaicing by Corner Matching

A robust image mosaicing method is put forward to construct panoramic image.The method uses Harris corner detector to extract corner-points for each image.A normalized cross correlation method is used to coarsely match corner-pairs between image pairs.Due to the existence of the outliers,direct use of least square algorithm would lead to the inaccurate parameter estimation of the transformation model that is required in image registration.To solve the problem,the Random Sample Consensus(RANSAC) algorithm is introduced to estimate the transformation model parameters,which is able to eliminate the outliers and to obtain exact corner-pairs.According to the aforementioned steps,the whole process of the image mosaicing is automatically accomplished.Some real images are subject to corner detection,registration and final mosaicing,which results in an excellent outcome.