Improved convolutional neural networks for age and gender prediction

Abstract

Age and gender classification is becoming significant to a considerable amount of software. Age and gender classification is a very interesting task because it’s a fine-grained classification image in which miniature distinctions may make the discrepancy in the predictions. However, age and gender classification is challenging problem because age and gender are affected by many factors negatively or positively. Up to now, there has been a considerable amount of researches on this subject. Convolutional neural networks (CNNs), which are one of the most ubiquitously used deep learning method, are adopted to enhance significantly performance in age and gender classification. This thesis presents a CNN-based method to improve accuracy of age and gender classification of both single-face and multi-face input image. The proposed method is based on a neural network architecture by which all of the experiments for both age and gender classification are applied. Two challenging datasets are trained to evaluate the accuracy of the proposed method. Keywords: age and gender classification, convolutional neural network, deep learning, multi-face input image.