Attention Mechanism Based Convolutional Neural Network for Unbiased Facial Recognition
Download as PDF
Haoyang Zou, Jianbo Chen
Facial recognition is an important and topical task in the field of artificial intelligence. In the last decade, a lot of algorithms have been developed to solve facial recognition (FR) problem. However, the prevailing facial recognition datasets are unbalanced in terms of age, gender, and race. Therefore, most of the previous algorithms trained on those datasets are biased. What’s more, most of the previous methods lack the ability to efficiently focus on the salient regions of a facial image. To solve the above-mentioned problems, a much more balanced dataset containing people from different ages, races and genders are collected, and an attention based method is proposed to automatically focus on the most important and salient local regions. Extensive experiment has been conducted, and the results demonstrate the effectiveness of the proposed method.
Attention mechanism, Neural network, Unbiased face recognition