Dual Stage Augmented Colorful Texture Synthesis from Hand Sketch

In this paper, we investigate the texture synthesis method generated from the hand-made sketches. In recent years, GANs have been vigorously studied in the field of image synthesis and generation, yet the texture synthesis from the hand sketch has not been extensively studied. In order to enable the synthesized image not only to possess the texture features, but also to show vibrant colors, we propose a cascaded network model that can synthesize a texture image. The proposed framework firstly generates a grayscale image with basic texture properties from hand sketch based on the conditional GANs. This grayscale texture is then colorized in the second stage. The network in the second stage is pre-trained using our constructed dataset to learn how to translate the grayscale image to a colorful image. We design a series of experiments to validate the effectiveness of our method. Encouraging results are achieved. The results demonstrate that the dual stage model outperforms the state-of-art generative models in the related areas.