一种具有空间约束的快速神经风格迁移方法  

Method of Fast Neural Style Transfer with Spatial Constraint

作　　者：刘洪麟 帅仁俊 LIU Hong-lin;SHUAI Ren-jun(College of Computer Science and Technology,Nanjing Tech University,Nanjing 211816,China)

机构地区：[1]南京工业大学计算机科学与技术学院,南京211816

出　　处：《计算机科学》2019年第3期283-286,共4页Computer Science

摘　　要：在神经风格迁移(Neural Style Transfer)技术中常用Gram矩阵进行图像风格提取,简单来说就是将各个特征进行内积。这样Gram矩阵只能提取其静态特征,对图片中物体的空间序列完全没有约束。文中提出了一种具有空间约束的快速神经风格迁移方法。首先,使用残差重新设计了快速神经风格迁移的转换网络。然后,运用空间偏移的方法对Feature map(特征图)进行变换,让变换后的Feature map T(al)进行Gram矩阵计算可得到相邻位置的互相关性。此互相关性包含空间信息,即约束了物体的空间序列。最终,实验表明了该方法对空间信息的约束明显优于传统方法,可以得到效果更好的风格化图像。Gram matrix,a method to get the inner product in simple terms,was commonly used for image style extraction in the style-transfer techniques.The Gram matrix can only extract the static features,but it is completely unconstrained to the spatial sequence of objects in the picture.This paper proposed a fast neural style transfer method with space constraints.First,the residuals are used to redesign the transform network of fast neural style transfer.Then,the method of spatial offset is used to transform the Feature map.Feature map T(a l ) are used for Gram matrix computation to get the cross-correlation,which contains the spatial information.That is to say,it can constrain the object’s spatial sequence in the picture.Finally,experiments show that the method’s ability of space constraint is better than traditional method,and the stylized image with better effect can be quickly obtained.

关 键 词：快速神经风格迁移 残差 GRAM矩阵 空间偏移 互相关性 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]