Gender recognition from torso features using elliptic fourier descriptors

Gender recognition has important applications in apparel design, social security, and human-computer interaction systems. In this paper, we investigate gender-recognition technologies using 3-D human body shape. The front and side silhouettes from 459 female subjects and 107 male subjects were extracted and then modeled using normalized Elliptic Fourier descriptors. Principal Component Analysis (PCA) was conducted to summarize the information contained by the EF coefficients. A back propagation (BP) neural network with 33 inputs, 2 outputs and 10 hidden layers was adopted to gender recognition. The research demonstrates that the gender recognition from torso features has achieved a considerably high recognition rate. Moreover, the combination of the PCA and BP neural network have provided effective ways for gender recognition and overcome some limitations in other technologies.