Abstract: Eliminating false matching is an important part in image stitching technology. Traditional eliminating erroneous matching method in the field of image stitching is RANSAC algorithm, but this method need numerous iterations and complex computation, and it often can not completely eliminate the false matching. Focusing on these shortcomings in RANSAC, this paper presents an improved RANSAC algorithm which is based on the geometric constraints. Clustering and grouping the matching points, then establishing a prejudgment geometric constraints model with the two geometric relationship between each couple of matching points: （1）the slope of each connection segment of each couple is equal  and （2）the length  of each connection segment of each couple is equal, to pre-purify matching points. The experiments show that the algorithm compared to the traditional RANSAC algorithm, eliminates mis-matching, reduces the number of iterations, improves computational efficiency, thereby improves the efficiency of image matching algorithm.

Key words: image stitching, Random Sample Consensus（RANSAC） algorithm, geometric constraints, pre-purified

摘要： 图像拼接技术中消除特征点误匹配是一项重要环节，针对传统的消除误匹配的RANSAC算法迭代次数多，计算复杂度较大且不能完全消除误匹配等缺点，提出了一种基于几何约束的RANSAC改进算法。该算法将几何约束法应用到RANSAC算法中，对图像特征匹配点进行聚类分组，根据每条匹配点对连接线的斜率应该相等、长度也应该相等这两个几何关系建立预判断模型，对匹配点对集合进行预提纯。实验证明，该算法相较于传统的RANSAC算法，误匹配基本消除，迭代次数减少，计算效率提高，从而提高了图像匹配算法的效率。

关键词: 图像拼接, 随机抽样一致（RANSAC）算法, 几何约束, 预提纯