一种基于图像纹理的模板匹配算法  

Template-matching method based on image texture

作　　者：刘铁军[1] 史泽林[1] 于海斌[1] 

机构地区：[1]中国科学院沈阳自动化所,辽宁沈阳110015

出　　处：《红外与激光工程》2005年第3期319-323,共5页Infrared and Laser Engineering

基　　金：中国科学院创新基金项目(CX02-04-02)

摘　　要：为了提高实际工程应用中图像匹配算法计算的实时性,增强图像匹配算法对复杂场景跟踪的鲁棒性,提出了一种新的基于图像纹理的模板匹配算法(TTM)。依据邻域内像素灰度变化的趋势分别提取水平、垂直方向二值化的图像纹理矩阵,然后根据定义的一个相似性判别准则,分别度量水平、垂直两个方向二值化图像矩阵之间的相关置信度。最后,通过合成两个方向上的图像匹配置信度结果得到目标配准置信度。实验结果表明,该算法对图像光照变化、目标部分被遮挡等情况具有较好的适应性。To improve real -time performance of image registration and enhance robust performance of target tracking under the complex scene, a novel template -matching method based on image region texture is presented. Binary texture matrixes of horizontal and vertical direction were constructed according to gray direction of neighborhood pixels, and then the similarity measurement criterion was defined to measure correlation credibility between binary texture matrixes of each dimension. The correlation credibility of appointed object was obtained through combination of two-direction registration result at last. Experiments prove that the proposed algorithm is robust for image illumination variance and target partial occlusion.

关 键 词：纹理特征 模板匹配 相关置信度 相似性度量 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]