基于方差约束与几何不变特性的图像匹配算法

为解决较多图像匹配算法主要通过测量关键点之间的距离来实现特征匹配，忽略了图像的结构信息，使其存在较多的误匹配的问题，本文设计了方差约束耦合几何不变特性的图像匹配算法。借助于Forstner算子计算像素点的兴趣值，以检测图像的特征；计算图像的梯度信息，获取图像的方向值，并切割图像特征的圆形邻域，从而获取扇形子域；以图像的方向值为基础，通过计算扇形子域中的灰度不变矩，输出对应的特征向量；引入区域方差函数，获取图像的结构信息，将其加入至图像特征的匹配过程，以约束欧式距离的测量结果，实现图像特征匹配；最后利用匹配点间的几何不变特性，对匹配特征去伪求真，优化匹配结果。测试数据表明：相对于已有的匹配技术，在对无变换图像、缩放图像以及旋转图像匹配时，所提算法拥有更高的匹配准确度，分别达到了96.56%、95.38%和93.52%。

Image matching method based on variance constraint coupled geometric invariance

Currently a number of image matching algorithms focus on measuring the distance between key points while ignoring the structure information of the images, hence are prone to mismatches. This paper presents an image matching algorithm using the geometric invariance of variance constraint coupling. With the help of Forstner operator, the interest value of pixels is calculated to detect the characteristics of the image. The gradient information of the image is calculated to obtain the direction value of the image. The circular neighborhood of image features is cut to obtain the fan-shaped sub domains. Based on the direction value of the image, the feature vector of the image feature is obtained by calculating the gray invariant moment in the fan-shaped sub domains. The region variance function is introduced to obtain the structure information of the image, which is added to the image feature matching process to constrain the results of Euclidean distance measurement and realize the image feature matching. Based on the geometric invariance between matching points, the matching image features are processed to get accurate image matching results. Experimental results show that compared with the existing matching techniques, this algorithm has higher matching accuracy, which up to 96.56%, 95.38% and 93.52% for non-transformed images, zoomed images and rotated images, respectively.