Residual Network for Face Progression and Regression

In computer vision application, the style transfer is a most active area, where deep generative networks have been used to achieve desired results. The development of adversarial networks training produces a high-quality image result in terms of face age progression and regression that is face aging and de-aging. Inspired by Ian Goodfellow, in this paper, we have designed the combinational network using the residual block, convolution and transpose convolutional in CycleGAN for face age progression and regression. Face aging is an image to image translation concept which is used in many applications such as cross-age verification and recognition, entertainment, in smart devices like biometric system for verification purpose etc. The proposed architecture preserves the original identity as it is and converts young people to old and vice versa. The network consists of residual blocks to extract deep features. The UTKFace unpaired image dataset is used to do experiments. The qualitative analysis of proposed methods in terms of performance metrics which gives better results. The performance metrics calculated such as Mean Squared Error (MSE), Root Mean Squared Error (RMSE) and Structured Similarity Index (SSIM) to the quality of image.