一种不变的基于傅立叶变换的区域形状描述子

An Invariant Region-Shape Descriptor Based on Fourier Transform

提出了一种不变的基于傅立叶变换的区域形状描述子.该方法首先将区域形状图像用极坐标光栅重新采样成形状像素矩阵,然后对该矩阵进行两次一维傅立叶变换和一次相位修正操作,最后用低频傅立叶系数构成的矩阵作为形状描述子.该形状描述子不仅保留了傅立叶系数中的相位信息,而且满足不变性.实验结果表明该方法要优于现有的傅立叶区域形状描述方法和Zernike矩形状描述方法.

A novel invariant region-shape descriptor based on Fourier transform is proposed in this paper. In this method, a shape pixel-matrix is firstly generated by image re-sampling using polar raster. Two times of 1D Fourier transforms and a conecting operator of phase are then conducted against the shape pixel matrix. The resulting matrix formed by the Fourier coefficients of low frequency is used to describe the region shape. The obtained shape descriptor not only preserves the phase information of the Fourier coefficients, but also achieves invariance to translation, scale and rotation. The experimental results show that the proposed method outperforms the existing Fourier region-shape descriptor and Zemike moment descriptor.