ABSTRACT
This chapter presents a model for generating thermal face identification by synthesizing visible pictures with the variational autoencoder model. This model successfully conserved the original face structure while simultaneously generating high-quality images from thermal data. In addition, in order to identify the images that have been generated, we make use of three different models that are based on lightweight architecture. This particular architecture has a fundamentally high accuracy in face detection because it reduces the number of parameters and executes at a high speed. After being trained and tested with a variety of datasets, the findings show that the proposed model is able to recognize thermal facial pictures. Because of this capability, the method may be used to new thermal face photos without the requirement for an additional training phase in order to identify the faces. The performance comparison reveals that the recently developed model is significantly more effective than the state-of-the-art approaches that are already in use, particularly with regard to the accuracy of scoring. In the future, generative adversarial networks (GAN) models may be included to the system. These models have the potential to narrow the gap between the synthetic and original pictures, which will result in an improvement in the image quality.
