Published Paper Details:

Face authentication is one of the embryonic issues for researches in modern days as it has many challenges and is used in several applications. The challenge of face recognition encounters large side angles in profile images as features between profile images and frontal images have large differences that lead to human identification. This is a reason much state-of-the-art facial recognition has reduced performance on human identification. In this research paper, we propose frontal face generation from profile face image before facial recognition using Global Generative Adversarial Networks (G-GAN). By using G-GAN along with unique error loss functions during training leads to good identity preservation and photo-realistic frontal face images. The Discrete Wavelet Transform (DWT) is used on generated frontal images from profile images and the frontal image stored in the database to extract features. The compressed low dimensional significant features from the low-frequency band are considered for face recognition. The Euclidean Distance (ED) between the ground true database frontal image and the frontal face images generated by G-GAN are computed for face recognition. The experimental results based on qualitative and quantitative analysis using available public datasets viz., Multi PIE, Bosporus, head pose-invariant, and Indian female are conducted and achieved good results. The experiment results of the projected scheme are related to the other existing approaches show the superiority of our method.