基于改进HU不变矩的快速图像匹配算法

Fast image matching algorithm based on improved HU invariant moment

为了解决当前HU不变矩匹配算法计算量大、无法快速匹配,以及对大幅度旋转和尺度变化匹配精度低的问题,提出了一种基于改进HU不变矩的快速图像匹配算法。利用金字塔模型,成比例降低模板图像和待匹配图像的分辨率,并提出了一种改进的匹配搜索策略,优化起始匹配位置和搜索路径,减少模板遍历计算量。提出了一种构造多尺度模板的方法,并利用相似度度量和连通域算法最终实现目标准确定位与标记。实验结果表明:所提算法对灰度变化、平移、大幅度旋转和缩放具有一定的不变性,且对大分辨率图像,匹配速度大大提高,取得了良好的匹配效果。

In order to solve the problems of current HU invariant moment matching,that the calculation is too large to match quickly,and the precision of large rotation and scale change matching is low,a fast image matching algorithm based on improved HU invariant moments is proposed.The algorithm uses a pyramid model,which proportionally reduces the resolution of the template image and the image to be matched,and propose an improved matching search strategy,which optimizes the starting match position and search path,reduces the computational complexity of template traversal.The algorithm proposes a method for constructing multi-scale templates,and use the similarity measure and connected domain algorithm to achieve the target accurate positioning and marking.The experimental results show that this algorithm has certain invariance properties in grayscale variation,translation,large rotation and scaling,the matching speed is improved greatly for large resolution image,and a good matching effect is obtained.