Abstract: A rectangle feature extracting method is proposed.The Canny edge detecting is used to get the images’ edge for its sensitive performance on images’ texture.Then distance transform on the result of Canny edge detecting is applied，which can reduce the interference caused by the complex texture region on rectangle pattern feature extracting.By consequence，the convolution image between distance transform result and mask image is computed，then the convolution image is classified and the rectangle pattern features’ control coordinate is got.Finally the rectangles’ four vertexes coordinate through geometric constraints are solved.The method is proved to be effective and workable by experiments，and can achieve 1.3 pixels on features’ location on real images.

Key words: feature extracting, Canny edge detecting, distance transform, rectangle pattern

摘要： 提出了一种图像矩形模式特征提取算法。首先利用Canny算子对图像纹理敏感特性求取图像的Canny边缘;对Canny边缘结果计算其距离变换图，可有效减小纹理复杂区域对后续矩形模式特征提取的影响;然后对距离变换图进行掩膜卷积，对其结果进行聚类，获得矩形模式特征的中心;最后利用几何约束求解出矩形模式特征的四个顶点。实验证明了该算法有效可行，特征定位精度在真实图像上可达1.3像素。

关键词: 特征提取, Canny边缘检测, 距离变换, 矩形模式