作者：赵玉兰；王兴伟；黄敏；姜春风

第一作者单位：吉林农业科技学院网络工程系

摘要：当前解决图像匹配算法主要是依靠独立像素点的灰度特性来完成图像配准,当匹配图像存在较大的灰度差异时,易出现较多的误匹配点,使得算法的鲁棒性与匹配正确度不高。提出基于区域统计直方图耦合自适应规则的图像匹配算法。首先,利用积分图像的方式求取矩形区域内像素值,再采用Hessian矩阵行列式提取特征点,以提高算法的计算效率;然后,引入极坐标系,对特征点的邻域进行分割,利用像素点对应的微分不变矢量的特性,构建区域统计直方图模型,用以求取分割区域内所有像素点的微分不变量的统计直方图,生成特征描述子。将特征点对应的描述子向量进行欧式度量,借助自适应规则对门限值进行整定,完成特征点的匹配,以提高算法的匹配精度以及鲁棒性;最后,引入PROSAC算法,通过对匹配特征点之间的极限距离平方和进行评估判定错误匹配点,以剔除错误匹配点。仿真实验结果表明,与当前图像匹配算法相比,所提出的算法具有更高的配准精度与鲁棒性。该算法在图像伪造、图像检索等信息应用领域具有重要应用价值。

Abstract：At present, many image matching algorithms mainly rely on the gray characteristics of independent pixels to complete image matching, when there are different in grayscale image matching, image matching error points will appear more, which makes the algorithms' robust matching accuracy is poor and lack of decline. In this paper, an image matching algorithm based on the regional statistical histogram coupling adaptive rule was proposed. Firstly, the sum of the pixel values in the rectangular region is obtained by using the integral image, then the feature points are extracted by the Hessian matrix determinant to improve the computational efficiency of the algorithm;Then, the neighborhood of feature points segmentation is used to construct polar coordinates, and the pixels corresponding to the differential invariant vector are used to construct region histogram model.By taking the segmentation histogram of differential invariants of all pixels within the region, the complete feature descriptor is generated. The descriptor vector of the corresponding feature points of Euclidean metric is measureds, and the use of adaptive threshold rules are tuned to complete feature point matching, so as to improve the matching accuracy and the robustness of the algorithm;Finally, in order to achieve the effect of eliminating the false matching points, by using the PROSAC algorithm, the error matching points ae deter mined through the limit of the square of the distance between feature point matching and evaluation. The simulation results show that compared to the current image matching algorithm, the proposed algorithm not only has the advantage of better matching image, but also has better robustness. This algorithm has important application value in the fields of image forgery and image retrieval.

基金：国家自然科学基金资助项目(61572123,61502092)；国家杰出青年科学基金资助项目(61225012,71325002)；中国博士后科学基金资助项目(2016M591449)；

关键词：图像匹配;Hessian矩阵行列式;区域统计直方图;自适应规则;PROSAC算法;

DOI：10.16186/j.cnki.1673-9787.2017.06.019

分类号：TP391.41