Age and Gender Recognition with Random Occluded Data Augmentation on Facial Images

• Main Authors: Chia-Yuan Hsu, 許家源
• Other Authors: Chang-Hong Lin
• Format: Others
• Language: en_US
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/275fd8

碩士 === 國立臺灣科技大學 === 電子工程系 === 107 === Facial analysis tasks have always been hot topics over the years for their broad varieties of applications, such as surveillance, commercial uses, human-machine interaction, and entertainment, etc. With the recent success of deep learning, these facial analysis tasks were able to tackle more the difficult and practical situations. However, while some tasks like age and gender estimation did achieve substantial improvements to the traditional machine learning based methods, they are still far from perfect to satisfy the need of real-life applications. This thesis proposed a data augmentation method by altering the training images that resemble real-life photos to improve the performance of the networks by providing more varieties to the training samples. The proposed method, Random Occlusion, adopted three simple occlusion techniques, Blackout, Random Brightness, and Blur, each simulating a different kind of challenge that would be encountered in real-world applications. We verify the effectiveness of our proposed method by implementing the augmentation method on two convolution neural networks (CNNs), the modified AdienceNet and VGG16 to perform age and gender classification. The proposed augmentation method improves the age accuracy results of the modified the AdienceNet and VGG16 by 1.0% and 0.8%, respectively; and gender accuracy results of the AdienceNet and VGG16 by 1.5% and 1.2%, respectively.