Gender Classification using Multi-Level Wavelets on Real World Face Images

Gender classification is a major area of classification that has generated a lot of academic and research interest over the past decade or so. Being a recent area of interest in classification, there is still a lot of opportunity for further improvements in the existing techniques and their capabilities. In this paper, an attempt has been made to cover some of the limitations that the associated research community has faced by proposing a novel gender classification technique. In this technique, discrete wavelet transform has been used up to five levels for the purpose of feature extraction. To accommodate pose and expression variations, the energies of sub-bands are calculated and combined at the end. Only those features are used which are considered significant, and this significance is measured using Particle Swarm Optimization (PSO). The experimentation performed on real world images has shown a significant classification improvement and accuracy to the tune of 97%. The results also reveal the superiority of the proposed technique over others in its robustness, efficiency, illumination and pose change variation detection.