基于Zernike矩的图像归一化技术的研究.pdfVIP

第30卷　第12期 计　算　机　工　程 2004年6月 Vol.30 № 12 Computer Engineering June 2004 ·基金项目论文 · 文章编号：1000—3428(2004)12 —0034—03 文献标识码：A 中图分类号： TP391.41 基于Zernike矩的图像归一化技术的研究 杨政武，方 涛 (上海交通大学模式识别与智能系统研究所，上海200030) 摘 要：推导了基于Zernike矩的图像归一化标准，实验分析了数字图像离散化和噪声对Zernike矩计算的影响，并提出了相应的改进措施。最 后，以 神经网络作为分类器检验了改进的 矩描述子的性能。试验结果显示改进的 矩描述子能非常有效地保持平移、尺度、 BP Zernike Zernike 对比度和旋转不变性，并且能够有效抑制噪声影响。 关键词： 矩；图像归一化；不变性；离散化；噪声； 神经网络 Zernike BP Research on Image Normalization Based on Zernike Moment YANG Zhengwu, FANG Tao (Institute of Pattern Recognition and Intelligent System, Shanghai Jiaotong University, Shanghai 200030) 【Abstract】In this paper, the ZM-based image normalization criteria are derived, and it experimentally analyzes the accuracy of the ZM-based image normalization under the circumstance of the discretization of digital image and the presence of noise, and accordingly proposes several approaches to improve the accuracy of ZM-based image normalization. The BP neural network is utilized as the classifier to test the performance of the improved ZM descriptor.The experimental results show that the improved ZM descriptor can near-perfectly preserve the invariance of translation, scale, contrast and rotation, and can efficiently restrain the influence of noise. 【Key words 】Zernike moment (ZM); Image normalization; Invariance; Discretization; Noise; BP neural network 形状描述子可以分为两类：基于轮廓的和基于区域的。