一种基于Radon变换的物体几何变换不变矩分析方法

An Approach to Shape Recognition Based on Radon Transforms

几何形状的识别在计算机视觉中具有重要意义,不变矩特征由于其在图像平移、伸缩、旋转时均保持不变,而且具有全局特性,是几何形状识别的主要方法。在已有的不变矩分析方法基础上,本文提出一种基于Ra-don变换的不变矩提取算法,用于对物体的几何变换不变性分析。该算法首先对图像进行坐标变换与归一化处理以实现平移与尺度变换不变,然后利用Radon变换将经过坐标变换与归一化处理后的目标图像转换到Radon投影空间,组成投影矩阵,再从投影矩阵中提取不变矩A(r)、E(r)进行目标图像的识别与分类。理论分析与实验结果表明,与现有的不变矩分析方法相比,该算法对噪声的鲁棒性强、时间复杂度低,仅用有限的几个矩即可以达到很好的分类效果。

Shape recognition is a very important issue in computer vision and digital image processing. Invariant moment has extensive applications in the field of Shape recognition due to its ability to represent global features and characteristics independent of translation, scale and rotation. This paper has addressed the issue of selecting invariant features for shape recognition in translation, scaling and rotation. Scaling and translation of an image were firstly normalized by coordinate transform and then the Radon transform was applied to the results to obtain a projection matrix. Finally, the set of invariant moments A（r） ,E（r） was derived from the matrix. Theoretical and experimental results show that the superiority of this approach includes low computational com- plexity and high robustness to additive noise in comparison with some recent methods. The results also show that the proposed approach can achieve better classification performance by a few descriptors.