Treffer: Single stage GAN-gabor wavelet fusion model for facial expression synthesis.

High-quality, photo-realistic image synthesis has significantly advanced since the development of Generative Adversarial Network (GAN), and the same is evident for facial expression synthesis (FES). FES finds application in the fields of computer vision, computer graphics, human-computer interface (HCI), etc. Deep learning based-architectures for FES have multiple stages of training, with a huge number of calculations and trainable parameters. Multiple stages of training can lead to higher training time and inter-dependency between stages, both makes the training process complex. In this work, we propose a novel approach using Gabor wavelets to extract facial features and a GAN architecture with a single stage of training. To enhance the receptive field of the feature extraction process, dilated convolutions are utilized. This enables the extraction of both global and local features, resulting in a more comprehensive representation of the input data, with lesser computations. Additionally, the modified architecture is superior in terms of reducing the computational complexity of the network, better image quality, and higher training speed. In the proposed model, the Gabor wavelets capture high-dimensional facial features in a specific orientation from the pre-processed face images. The proposed model is evaluated using a Convolution Neural Network (CNN) based expression classifier; Principal Component Analysis (PCA) based face classifier and FID (Frechet Inception Distance) score. Comparative results demonstrate that the proposed model converges quickly with fewer training parameters and produces higher-quality facial images. [ABSTRACT FROM AUTHOR]

Copyright of Signal, Image & Video Processing is the property of Springer Nature and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)