Our paper of Smart Interaction and Game Technology (SIGT) Lab is published in Biomedical Signal Processing and Control (BSPC) (IF: 5.067) [LINK]

Title: 

A Hybrid Approach of Neural Networks for Age and Gender Classification through Decision Fusion

Authors:

James Rwigema, Joseph Mfitumukiza, KimTae-Yong

Abstract:

Age and Gender are identified as very important attributes in human identification and these attributes are used in various fields of Human and Computer Interaction (HCI) such as security systems, video-surveillance systems, online purchasing systems, judicial systems, transport, medicine, and so many others. In recent years, age and gender estimation based on facial feature analysis have been articulated as a challenging research topic by many researchers in the HCI field. In this research, we aim to present a novel hybrid algorithm of Conventional Artificial Neural Networks (C-ANN) and Convolution Neural Networks (CNN) by using decision fusion techniques for age and gender classification. The novelty of our research is the fusion of the decisions obtained by the two neural networks to increase the accuracy of age and gender estimation. We used the probabilistic decision fusion techniques such as majority voting decision fusion, Naïve-Bayes combinationdecision fusion and sum rule decision fusion for better recognition accuracy rate. Among these techniques, the sum rule decision fusion provided the highest accuracy rate compared to the state of art because of reducing the adjacent classes’ likelihoods during decision classifications.