Effect of Image Distortion on Facial Age and Gender Classification Performance of Convolutional Neural Networks
Choon-Boon Ng, Wei-Haw Lo · IOP Conference Series Materials Science and Engineering · 2019
Significant improvement in the task of age group categorization and gender classification of facial images has been achieved using deep convolution neural networks (CNN). In this paper, we study the effect of image distortions such as blur, noise, rotation and occlusion on the performance of a state-of-the-art CNN. We found that the CNN was more sensitive to noise compared to blurring, especially for age estimation. By studying occlusion, we also identified the salient regions of the face. An interesting result is that the upper half of the face is more important for age estimation, while for gender classification it is the lower half. These insights should prove useful for future development of CNN models for facial age and gender classification.