中文摘要：

为克服单点描述子匹配数量少、匹配正确率低等问题,提出一种三角组合约束下的尺度不变特征描述子.首先利用几何约束将满足条件的3个特征点组合为三角形;然后利用三角形内切圆半径作为支撑区域确定的依据,对获取的三角形构建尺度不变的特征描述子并进行匹配;最后根据支撑区域主方向信息将三角形匹配转换为点匹配,并利用重复匹配出现的概率去除错误匹配.实验结果表明,该方法不仅对旋转、尺度变化、视角变化、JPEG压缩等图像变化具有鲁棒性,而且匹配的特征点数量多、匹配准确率较高.

英文摘要：

In this paper, we propose a scale-invariant feature descriptor under triangular combination constraint to overcome the issues of less number of the matched points and low matching accuracy for point descriptor. First,three feature points under given geometric constraints are combined into a triangle. Then, using the radius of the inscribed circle of the obtained triangle as the support region size, a scale-invariant descriptor for the triangle isthus created and the matching is made subsequently. Finally, the point correspondences are obtained by the computedmain direction of the support region, and mismatches are eliminated by retaining the most repeated pointmatch. The experimental results show that this method is not only robust to various image transforms, such as rotation,viewpoint change and JPEG compression, but also increases the number of the matched points with high accuracy.