一种利用邻近特征点夹角一致性约束的匹配提纯方法

Matched Points Purification Method Based on the Consistency Constraint of Adjacent Feature Points' Angles

基于特征点的匹配是图像检索领域的重要研究内容,如何滤除误匹配点对更是研究的热点问题.针对特定目标对称性强、相似特征点多时容易产生误匹配的情况,提出一种利用邻近特征点夹角一致性约束的匹配提纯方法,并应用于车型识别.该方法根据两幅待匹配图像中对应邻近特征点夹角应当一致的原则,并利用局部图像块直方图信息,对SIFT特征点匹配结果提纯.实验表明该方法有效提高了匹配点的准确率,对图像旋转、缩放、角度变换等仿射变换有良好的鲁棒性,对图像模糊、光照、马赛克等影响也有一定抗干扰能力.

Key points based image matching is a fundamental problem in image retrieval,in which how to remove false matching point pairs is a hot issue in recent years. Aiming at such mismatch problems derived from high symmetric structures and lots of local similar key points in specific objects, a novel matching purification method based on angles constraint among key points and their neighbors is proposed and then applied to vehicle type identification. This method adopts the rule that the neighboring angles of matching point pairs in two images should be consistent. And the local region＇s histogram information is also considered to purify matching results for SIFT key points. Experiments show that the proposed method improves the accuracy of key points＇ matching, which is robust to affine transformation, such as rotation, scaling, and view point change, and has certain anti-interference ability to image blur, illumination var- iation and mosaic.