Data-driven Automatic Facial Image Colorization using an Encoder-Decoder Network

Facial image colorization is a challenging task in computer vision, aimed at generating realistic and accurate colored versions of grayscale facial images. This study introduces a novel approach that leverages the power of an encoder-decoder network to predict and generate high-quality colored facial images from their grayscale counterparts. The proposed method surpasses other state-of-the-art techniques, showcasing promising and superior results. By harnessing the representation learning capabilities of the encoder-decoder network, meaningful facial features are captured, enabling effective inference of appropriate color distributions. This architecture facilitates realistic and visually appealing colorizations while preserving intricate facial details. Comprehensive visual comparisons with other state-of-the-art methods demonstrate significant improvements in color fidelity and overall image quality. Experimental results validate the efficiency of the encoder-decoder network in accurately predicting facial colors, even in challenging regions such as hair, eyes, and lips. The colorized images exhibit vibrant and lifelike tones, providing a faithful representation of the original grayscale images.