Generic ring fourier descriptors for shape description

The traditional Fourier descriptor (FD) can only be used to objects with single boundary. We develop a novel method, Generic Ring Fourier Descriptor (GRFD), which can be applicable to gray-scale image. Firstly, we take an unconventional view of the traditional FDs, as we treat the boundary as a closed-curve in a 2-D binary image. Consequently, the generated FD (GFD) is proposed such that it can be applicable to a closed-curve with gray value in an image. Then GFD is applied to concentric circles with different radius in the image and GRFDs are derived. Objects can be represented by these GRFDs which are invariant to translation, scaling and rotation. Experiments using images from the coil-20 database have been conducted to compare the performance of the proposed descriptors with Hu's moments and Complex moments (CMs). The proposed method demonstrated superior performance.