Human Shape Recognition using Fourier Descriptor

The aim of this study is to investigate Fourier Descriptor (FD) as feature vectors for shape representation and recognition since FD is the best known boundary based shape descriptor and has proven to outperform most other boundary based methods in terms of accuracy. Furthermore, FD is also invariant to geometric transformations and has good noise tolerance. The main concern regarding FD is the number of terms that need to be maintained from the original Fourier transform for effective representation and description. A system that computed FDs of human and non human from their silhouettes; normalized the descriptors and further applied as feature vectors for recognition is developed. Initial results of experiment showed that using adequate number of both low and high frequency components could represent the shape based on high recognition rate achieved. The process of shape recognition using FDs looks promising.