【摘要】 随着人工智能技术的提高，人们对图片处理的要求也越来越高，对于图片风格的迁移，也越来越多样化和艺术化，传统的图片风格迁移已满足不了人们现在的需求，基于Gatys等的风格迁移针对艺术化的风格图片的迁移效果明显，但是采用真实化的风格图片其迁移后的图片存在局部空间扭曲，效果不是很好，因此针对上述缺点，论文在Gatys等风格迁移算法的基础上改进了损失函数，在损失部分引入了正则损失，最大程度保留了图片的细节，同时也对风格损失函数进行了改进，减少了风格溢出情况。经测试，迁移的结果无论是在色彩表现上还是纹理特征上都具有很好的迁移效果，与Gatys等算法相比，生成图片的峰值信噪比（PSNR）提高了近3.2%，结构相似性（SSIM）提高了近3%。更多还原

【Abstract】 With the development of artificial intelligence technology,the requirements for image processing have been continuously improved,and image style transfer has become increasingly diverse and artistic. Traditional image style transfer methods can no longer meet current practical demands. The style transfer method proposed by Gatys et al. achieves remarkable effects on artistic style images,but it leads to local spatial distortions when applied to realistic style images,resulting in unsatisfactory transfer results. To solve the above problems,this paper improves the loss function on the basis of the style transfer algorithm by Gatys et al.Regularization loss is introduced into the loss part to preserve image details to the greatest extent,and the style loss function is optimized to reduce style overflow. Experimental results show that the transfer results have good performance in both color representation and texture features. Compared with the algorithm of Gatys et al.,the peak signal-to-noise ratio(PSNR)of the generated images is increased by nearly 3.2%,and the structural similarity(SSIM)is improved by nearly 3%.更多还原