【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。更多还原

【Abstract】 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 circs that 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.更多还原