中文摘要：

在图像拼接中,误配点剔除是关键的一步.为了提高误匹配点的剔除效率,提出了一种快速的误配点剔除算法.算法针对图像序列特点,在随机抽样一致算法的基础上,首先对匹配点进行排序,采用分段随机法选点以及建立变换矩阵;其次运用区间限定的预检测模型,对通过检测的变换矩阵进行全局检验得到内点集;最后使用最小二乘法修正变换矩阵.与传统算法相比,该算法具有更低的时间复杂度、更高的计算精度和更好的稳定性,在数据质量较差情况下效果更加明显.

英文摘要：

The elimination of mismatching point is a key step in image mosaic.To speed up the mismatching point elimination in image matching,an efficient algorithm for mismatching point elimination is presented.Based on the RANSAC（random sample consensus） algorithm,the algorithm aims at the feature of the image sequence.First,the match points are sorted,divided into three parts and piecewise picked randomly to estimate the transforming matrix.Second,the matrix is cursorily checked with a novel interzone limited pre-test model,a further check is executed on the matrix that passes the pre-testing to get inliers.Finally,the least-square method is used to the inliers to get the real matrix.Experimental results suggest that the proposed algorithm is of lower complexity,higher accuracy and stableness especially in cruel conditions,which meets the demand of image sequence mosaic well.