结合PCA和FPD的傅里叶形状描述子

摘要

This paper proposes a new Fourier shape descriptor integrated with Principal Component Analysis(PCA) and the Farthest Point Distance(FPD). It gets the canonical form of the shape by PCA, and the FPD is extracted from the resampling dataset. It uses the vector of FPD for the Fourier Transform(FT). After above processing, the descriptors are insensitive to the affine transform, rotation, noise and so on. They can hold the local features well, such as corners. By experimenting on the popular shape dataset, it gets a promising recognition rate of 90.6% and 93% for hand-tools and affine transformed dataset, respectively. It proves the method is effective.%提出一种将士成分分析(PCA)与最远点距离(FPD)相结合的二维形状傅里叶描述子.利用PCA方法对二维形状进行归一化,对归一化后的数据进行重采样,提取采样点的FPD用于傅里叶变换,运用得到的描述子对二维形状进行分类.经该方法处理得到的描述子对仿射、旋转、噪声等的抗干扰能力得到提升,对形状的局部特性描述能力更强.在手工工具数据集和仿射变换数据集中进行测试,结果表明,该方法的识别率分别达到90.6％和93％,从而验证其有效性.