基于卷积神经网络的图像风格迁移算法研究

Research on Image Style Transfer Algorithm Basedon Convolutional Neural Networks

针对Gatys的图像风格迁移算法做了两个方面的改进,首先提出了一种更加适用于风格迁移的卷积网络结构,相较于其他的预训练卷积神经网络模型减少了95%的参数数量,降低了22%以上算法运行时间;其次对风格迁移的风格损失函数部分做了改进,可以使一幅内容图像同时迁移多种不同的画作风格。

Two improvements are made to Gatys’image style transfer method.First,a convolutional network structure that is more suitable for style transfer is proposed.Compared with other pre-trained convolutional neural network models,the number of parameters is reduced by 95%,More than 22%of the algorithm running time.Secondly,the style loss function of style transfer has been improved,so that a content image can transfer multiple different painting styles at the same time.