区域灰度分布耦合相似判定策略的图像匹配算法

Image Matching Algorithm Based on Regional Gray Level Distribution Coupling Similarity Determination Strategy

目的为了解决当前因图像匹配算法主要依靠提取图像的特征属性矢量进行度量,从而利用其对应的相关系数最大的点进行匹配时导致匹配结果中存在较多的错误匹配点以及匹配误差变大的问题。方法提出区域灰度分布耦合相似判定策略的图像匹配算法,首先利用Forstner算子来提取图像的特征点,以特征点为中心,采取建立极坐标系的方法来确定特征点的主方向,通过特征点邻域的灰度特征来生成低维度的特征描述子;然后引入归一化互相(NCC)函数对特征点之间的相似度进行评估,建立矩形窗口特征点双向匹配规则,完成特征点的匹配,以提高特征点之间的匹配准确度和算法鲁棒性;最后,根据正确匹配特征点组成的三角形具有相似性的特征,设计相似判定策略,对错误匹配点进行剔除,以改善匹配精度。结果实验结果表明,与当前图像匹配技术相比,文中匹配算法具有更高的匹配精度与效率,有效降低了特征点的误匹配率。结论所提图像匹配技术具有较高的配准精度,在图像伪造、包装条码识别等领域具有一定的应用价值。

The work aims to solve the problem that the measurement is currently done mainly by depending on the extraction of feature vector, thus causing many mismatching points in the matching results and the increasingly bigger matching errors during the matching by means of its maximum point of the corresponding correlation coefficient. An image matching algorithm based on the regional gray level distribution coupling similarity determination strategy was proposed. First of all, the feature points of the image were extracted with the Forstner operator. With the feature points as the center, the main direction of the feature points was determined by establishing the polar coordinate system. The feature descriptor was generated through the neighborhood gray feature of the feature points. Then, the similarity between the feature points was evaluated by introducing the normalized cross correlation（NCC） function. The bidirectional matching rule of a rectangular window feature point was established to complete the feature point matching, in order to improve the matching accuracy and robustness of the algorithm between the feature points. Finally, a similarity determination strategy was designed based on the similarity of triangles formed by accurately matched feature points, which was used to eliminate the mismatching points, so as to improve the matching accuracy. The experimental results showed that, compared to the current image matching technology, the matching algorithm herein had higher matching accuracy and efficiency, which effectively reduced the mismatching rate of the feature points. The proposed image matching technology has higher registration accuracy and provides certain application value in such fields as image forgery and pack-aging bar code recognition.