Treballs Finals de Grau d'Enginyeria Informàtica, Facultat de Matemàtiques, Universitat de Barcelona, Any: 2016, Director: Santi Seguí Mesquida
Colorizing is the act of giving color to grayscale images. A convolutional-neural-network-based method to colorize images without human interaction is presented in this project. Various frameworks, architectures, color spaces and approximations are explored to obtain the final model, capable of correctly restoring the original color of photographies without any further information than the image itself.
The principal aim of this project is to propose an idempotent architecture that could be trained with all kinds of images and yet produce good results. To demonstrate how the process works and show the obtained results, three categories of images will be used along this project: synthetic images representing numbers, landscape images and human faces.