摘要
提出了一种基于 Radon变换的新的图像识别方法。首先构造了二维图像在 Radon变换空间的平移和比例矩不变量 ,基于该矩不变量引入整体平均技术获得了更好的抗噪效果 ,并根据变换空间数据的特性 ,提出了利用奇异值分解得到一般意义下的旋转不变量。利用不同的图像库进行识别研究 ,实验结果指出了两种方法的不同应用条件。
Refs.7 and 8 by Galigekere et al, in our opinion, still suffer from two shortcomings: (1) their image recognition method based on Radon transform is still imperfect in that they deal successfully with only scale invariant and translation invariant but they fail to be successful in dealing with rotation invariant; (2) their method is not satisfactory under noisy conditions. Our aim is to remove these two shortcomings. The procedure of the moment construction is given. Based on the moment in the Radon transform space, we make some improvements, such as: (1) ensemble average technique is introduced in calculating the Radon moment; (2) according to the data in the transform space, singular value decomposition (SVD) is provided here to obtain rotation invariant; (3) based on the invariant moment using SVD, a pattern recognition procedure is given. Experimental research on different sets of images was carried out. Experimental results do show preliminarily that we succeed in removing the above-mentioned two shortcomings. We must acknowledge that our method does lead to additional computational complexity.
出处
《西北工业大学学报》
EI
CAS
CSCD
北大核心
2004年第3期392-396,共5页
Journal of Northwestern Polytechnical University
基金
国家自然科学基金 (6 0 172 0 37)资助
关键词
RADON变换
矩不变量
奇异值分解
图像识别
Radon transform, invariant moment, SVD (singular value decomposition), image recognition
