SCGAN: Saliency Map-Guided Colorization With Generative Adversarial Network

Given a grayscale photograph, the colorization system estimates visually plausible colorful image. Conventional methods often use semantics to colorize images. However, in these methods, only classification semantic information is embedded, resulting confusion and color bleeding final colorized To address issues, we propose fully automatic Saliency Map-guided Colorization with Generative Adversarial Network (SCGAN) framework. It jointly predicts saliency map minimize Since global features from pre-trained VGG-16-Gray network are embedded encoder, proposed SCGAN can be trained much less data than state-of-the-art achieve perceptually reasonable colorization. In addition, novel map-based guidance method. Branches of decoder used predict as proxy target. Moreover, two hierarchical discriminators utilized for generated map, respectively, order strengthen visual perception performance. The evaluated on ImageNet validation set. Experimental results show that generate more images techniques.