Deep convolutional neural network based colorization
Abstract
Colorization is of great importance in restoring old gray pictures and making them more vivid. Thanks to the recent success of deep neural networks in various problems of computer vision, deep convolutional neural network has also been proposed for colorization and has brought about promising results. Yet previous works usually focus on some dataset and do not consider the influence of training data. This paper builds and evaluates many colorization models on various datasets according to criteria of image quality. The experimental results show interesting performance of residual networks in colorization. Moreover, an appropriate choice of training data may help to build an effective colorization model.
Moraref