Image Colorization Algorithm based on Self-Attention Network

In the task of colorizing gray image, it has been a challenging problem that the reconstructed color images have blur boundaries. In order to solve this problem, a novel method based on self-attention network is proposed in this work. In order to improve the color effect and increase the color contrast. An end-to-end deep learning model is constructed by using efficient convolution combination and self-attention network to extract image features, to learn the spatial dependence of features and the internal correlation between channels. In this way, the reconstructed performance can be improved with a better color effect and contrast. Using the pixel-wise color loss and the generative adversarial networks loss, the network parameters are optimized continuously to guide the generation of high-quality images. Compared with other state-of-the-art algorithms, the proposed method can result in color images with more clear boundary detail and more natural coloring effect than other approaches.