摘要
在常规图像风格迁移算法中,因缺少对于图像特定风格的特征对齐,图像风格迁移的速度较慢。因此,研究基于深度生成模型的图像风格迁移算法。通过分析图像风格的迁移,运用Sigmoid函数和ReLU函数实施函数的激活,在不同的卷积层中对图像特征进行风格对齐,基于深度生成模型改进算法;基于图像风格扩展采样点,对范围内的采样图像进行合成,实现对图像风格迁移算法的改进。验证实验表明,文章提出的算法平均迁移速率为10.26 MB/s,高于对照组,从实验结果可以看出,改进后的算法具有一定的速率优势。
In the conventional image style migration algorithm,the speed of image style migration is slow because of the lack of feature alignment for image specific style.Therefore,the image style migration algorithm based on depth generation model is studied.By analyzing the transfer of image style,Sigmoid function and ReLU function are used to activate the function,and the image features are aligned in different convolution layers,and the algorithm is improved based on the depth generation model.Based on the image style extension sampling points,the sampled images in the range are synthesized to improve the image style transfer algorithm.The verification experiment shows that the average migration rate of the algorithm proposed in this paper is 10.26 MB/s,which is higher than that of the control group.From the experimental results,the improved algorithm has certain speed advantage.
作者
陈洋
赵茹楠
CHEN Yang;ZHAO Runan(School of Information Engineering,Zhengzhou University of Science and Technology,Zhengzhou Henan 450064,China)
出处
《信息与电脑》
2023年第14期97-99,共3页
Information & Computer
关键词
深度生成模型
图像风格
迁移算法
算法改进
depth generation model
image style
migration algorithm
algorithm improvement