计算机应用 ›› 2010, Vol. 30 ›› Issue (2): 441-444.

• 模式识别 • 上一篇    下一篇

基于形状模板的快速高精度可靠图像匹配

赵振民1,彭国华2,符立梅3   

  1. 1. 西北工业大学;香港理工大学
    2. 西北工业大学
    3.
  • 收稿日期:2009-08-05 修回日期:2009-09-23 发布日期:2010-02-10 出版日期:2010-02-01
  • 通讯作者: 赵振民
  • 基金资助:
    国家自然科学基金资助项目

Fast high-precision reliable image matching algorithm based on shape

  • Received:2009-08-05 Revised:2009-09-23 Online:2010-02-10 Published:2010-02-01
  • Contact: ZHAO Zhen-Min

摘要: 为了提高工业检测中图像匹配精度和速度,提出一种基于形状模板的快速高精度图像配准算法:根据定义的图像匹配相似度量,采用图像金字塔搜索匹配策略,利用形状信息进行模板匹配。具体流程为:首先在参考图像上选择感兴趣区域生成模板,使用Canny滤波器对模板和搜索图像进行滤波,并计算边缘点的方向向量;其次,在此基础上构造该模板和搜索图像的图像金字塔,在图像金字塔最高层图像进行完全遍历匹配,获得具有匹配分值的潜在匹配点,然后根据匹配分值大小逐层逐次跟踪潜在匹配点,进行匹配,直至图像金字塔最底层;最后使用最小二乘法调整位姿参数,使其达到亚像素精度。实验表明该方法匹配速度快,匹配精度高,而且匹配鲁棒性高,不受遮挡、混乱、非线性光照变化、离焦、对比度低、全局对比度反转、局部对比度反转等情况的影响,完全可以满足实际工业需求。

关键词: 形状模板匹配, 图像金字塔, 最小二乘法, 方向向量, Canny算子

Abstract: A fast and high-precision reliable shape-based matching algorithm is proposed to improve industrial detection precision and speed of image matching in this paper. First, the definition of image matching similarity measure is proposed. Then, image search matching strategy was adopted from coarse to fine. Specific procedures are given. First, the template is generated by selecting the region of interest in the reference image. Then, it is the process of filtering the template and the search image using canny operator, and computing the direction vectors of their edges. The image pyramids are constructed for the filtered template and the filtered search image on this basis. The image matching is carried on at the highest level of the image pyramid under the similarity measures. After the potential matches have been identified, they are tracked through the resolution hierarchy in successive until they are found at the lowest level of the image pyramid. Finally, the subpixel-precision pose parameters are achieved through the least-squares adjustment. Experiments demonstrate that the algorithm meets the actual industrial demand. It is robust to the occlusion, the clutter, the nonlinear illumination, the defocused, the low contrast, the global contrast change, the local contrast change, and so on. The precision of the image matching is less than 0.1 pixel. The speed of the image matching is less than 160ms for 512pixel*512pixel image, less than 270ms for 1024pixel*1024pixel image, less than 420ms for 2048pixel*2048pixel image.

Key words: shape-based image matching, image pyramid, least-square method, direction vector, Canny operator