Gray level image colorization method based on generative adversarial network

Abstract

The invention discloses a grayscale image colorization method based on a generative adversarial network, and the method comprises the steps: firstly, selecting a quantitative color image group in a COCO image data set, carrying out the decoloring processing, making a training set, constructing a generative adversarial network architecture, enabling a generator model to complete the pre-training in the generative adversarial network architecture, and carrying out the image colorization. And then alternately training the discriminant model and the pre-trained generative model, adjusting parameters to obtain a trained model, and inputting test data into the model to realize gray level image colorization. Through the pre-training method and process of the generator, the training method and data set optimization are greatly improved, the training time is greatly shortened on the basis of ensuring the training quality and the generalization quality of the finally generated image, and the method has flexibility; and training and testing are carried out on a COCO data set by utilizing a U-Net thought, so that the defects that manual intervention is needed and fine coloring work of a large-size image pixel level is difficult to carry out in a traditional method can be reduced to a great extent.

Description

Technical field [0001] The present invention involves the field of deep learning and image generation technology, which specifically involves a gray image color method based on generating the network. Background technique [0002] Gray image color algorithm is a research hotspot in the field of digital image processing and computer vision. It has broad application prospects in black and white film and television data, animation sketches, ancient painting repair, and medical and aviation. The traditional coloring method is mainly divided into two categories: local color diffusion method and color transfer method based on reference images. Among them, the local color diffusion method needs to rely on artificially gives some calibration color pixels to give the global method through diffusion or change points to the overall situation. Image dyeing, however, this method requires people to provide initial color diffusion pixels and bring inconvenience to the colorful process; the colo...

AI Technical Summary

Problems solved by technology

[0007] The purpose of the present invention is to provide a gray-scale image color speech method based on generating adversarial networks to solve the problems of long time-consuming and poor flexibility existing in the current technology

Images