Colorization of Thermal Facial Images into Visible Facial Image Using RGB-GAN

Images captured by thermal cameras are independent of lighting conditions. However, it can be challenging for human examiners to identify thermal face photos. Facial recognition technology enables automatic identification or verification of individuals in digital images or video frames extracted from video sequences. There are multiple methods employed by facial recognition systems, but they typically involve comparing the features extracted from a specific image with those stored in a database. This technology finds application in various areas, including access control and identification systems. It is worth noting that facial features can exhibit unique characteristics specific to an individual throughout their lifetime. In this paper, the process of colorizing thermal facial images into the visible spectrum based on Cycle GAN is undertaken. There are many variations of the GAN but to translate or map from the one domain image into another domain image the cycle GAN fits with its application. CycleGAN aims to acquire knowledge of the relationship between two distinct image collections originating from separate domains, each possessing unique styles, textures, or visual attributes. The RGB-GAN which is proposed in this paper refers to the red, green, and blue channel generative adversarial network that individually takes the independent images in the thermal format to colorize in the independent domain channel network-merging three generated channel results combined to make one RGB-colored image. One more generator network involves identifying and comparing the result with the original visible colored image to give feedback to the network. At last after training the final model, the classification task involves generating and classifying the correct person out group of persons when the thermal image is given as input. The output includes face recognition accuracy of generated images, comparative analysis with protocols and state-of-the-art techniques.