基于新的风格损失函数与亮度信息的图像风格迁移算法

Image style transfer algorithm based on new style loss function and brightness information

针对传统图像风格化处理算法在进行图像风格化处理过程中,出现颜色信息丢失、笔触细节错乱以及边缘线条扭曲等情况,本文基于卷积神经网络的图像风格迁移算法进行了改进,通过综合考虑图像的亮度颜色信息特性,提出一种基于新的风格损失函数与亮度信息的图像风格迁移算法。首先,将内容图像进行色彩空间转换,让风格转移单独在亮度通道上进行,保留了内容图像的色彩信息。其次,把直方图损失加入到总的损失函数中,构造新的总损失函数,提高了风格化图像的稳定性。最后,把提取的内容图像边缘信息叠加到结果图中,达到改善风格化结果图边缘信息模糊不清的目的。实验表明,改进后的方法在笔触细节、纹理结构以及色彩空间上有更好的整体效果表现。

In view of the loss of color information, disordered stroke details, and distortion of edge lines in the traditional image stylization processing algorithm, this paper improves the image style transfer algorithm based on convolutional neural network, and comprehensive consideration of image brightness and color information characteristics.This paper proposes an image style transfer algorithm based on a new style loss function and brightness information.First, the content image is converted into the color space, and the style transfer is performed on the brightness channel alone, retaining the color information of the content image.Second, the histogram loss is added to the total loss function to construct a new total loss function, which improves the stability of the stylized image.Final, the edge information of the extracted content image is superimposed on the result image to achieve the purpose of improving the vagueness of the edge information of the stylized result image.Experiments show that the improved method has better overall performance in stroke details, texture structure and color space.