快速风格迁移  

Fast Style Transfer

作　　者：钱小燕[1] 肖亮[1] 吴慧中[1] 

机构地区：[1]南京理工大学计算机科学与技术系,南京210094

出　　处：《计算机工程》2006年第21期15-17,46,共4页Computer Engineering

基　　金：高等学校博士点学科专项基金资助项目(20020288024);南京理工大学青年学者基金资助项目(Njust200401)

摘　　要：提出一种基于样图的快速风格迁移方法。对经典Ashikhmin算法进行了改进,引入源图像和目标图像及其低分辨率图像的统计信息的L2距离,建立一种新的邻域一致性度量。对于重复候选点,以随机生成候选点的方法扩大搜索范围。同时采用双向扫描策略保证风格传输中纹元的视觉完整性。实验证明,改进算法能够快速有效地实现图像间的风格传输。A fast style transfer approach is proposed. A new neighborhood difference metric is built by improving the classical Ashikhmin approach and introducing the statistical characteristic belonging to source image, target image and their blurred images into neighborhood similarity metrics. In addition, the cites thai Ashikhmin＇s approach tends to create repeated candidates by producing extra candidates in random positions is solved. Meanwhile, visual integrality is maintained by applying the bi-directional scanning into style transferring. Enough experiments and results prove that the improved algorithm can realize style transferring between images fast and effectively.

关 键 词：非真实感 纹理合成 纹理传输 艺术绘制 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]