基于深层聚合结构网络的灰度图像彩色化方法

当前灰度图像彩色化方法普遍存在边界晕染、细节丢失和着色效果枯燥等问题。针对以上问题,提出了一种基于改进的深层聚合结构网络的灰度图像彩色化方法。将深层聚合结构网络引入图像彩色化领域中,且在传统网络基础上加入长连接,在缓解网络梯度消失问题的同时提升其特征利用率,从而提升算法模型对图像边界和细节的处理能力。另外,模型融合生成对抗网络结构,搭建判别网络,动态评价图片彩色化质量,缓解着色枯燥的问题。实验证明,该方法相比于传统彩色化方法,减轻了着色时边界漏色问题,还原了更多的图像细节,图像颜色更为丰富。

Gray image colorization method based on deep layer aggregation

Current grayscale image colorization algorithm generally has the problem of boundary blooming,loss details and single coloring effect.This paper proposed a gray image coloring algorithm based on improved deep layer aggregation structure network.To improve the image boundaries and details processing ability of the algorithm model,long connection participated into the traditional network,which also reduced the problem of the disappearance of the network gradient and improved the utilization of features.In addition,the model combined the generation of confrontation networks.This paper also built a discriminative network that the color quality of images dynamically evaluated and alleviated the problem of single-colored.Experiments show that compared with the traditional colorization algorithm,the proposed algorithm not only reduces the problem of boundary color leakage during coloring,but also restores more image details and enriches image color.