面向角度一维投影模板的图像匹配方法

One-dimension angle-oriented projection template matching

针对交迭图像模板匹配中存在的模板选择和匹配判别两个关键环节,提出一种一维模板匹配方法.通过基于灰度差分割的模板选择原则降低亮度差和噪声的影响,同时利用线性插值获得面向角度一维灰度投影,最大化模板中的图像边缘信息,从而加快模板匹配过程.在面向角度投影模板的基础上进行改进AD(Absolute Differences)指标的模板匹配判别,减少了因图像旋转带来的误匹配,并获得除匹配中心坐标外匹配图像相对参考图像的偏转角.通过不同情况下匹配结果在精度和速度上的分析,实验证明该方法应用在大部分真实交迭图像的匹配效果优异,同时运行速度能满足图像匹配应用的需要.

Refer to two key stages in template matching between two overlapping images： template selection and iterative matching, an one-dimension angle-oriented projection template matching method was proposed. The matching process firstly evolved a template selection discipline based on gray level difference segmentation to reduce the sensibility to image noise and illumination changes. After the segmentation, a well post intensity projection into angle-oriented array by linear interpolation was used to accelerate matching process and to maximum the employment of edge information. The method at last implemented an angle-oriented reformed AD （absolute differences） matching principle both to reduce miss-matches conduced by viewpoint rotation and to obtain rotation between reference and target other than correspondences coordinates. With aid of analysis of matching process respect to different cases under consideration of speed and precision, experimental results proved that the proposed approach works well over most overlapping images and is fast enough for real time applications.