摘要
文章针对Cubic与Lanczos插值卷积核放大图像的斜边伪波动性,提出了一种边缘优化的图像插值卷积核搜索算法。通过相位采样对插值卷积核离散化,构造边缘波动率优化的插值目标,基于带约束的非线性最小值优化方法数值求解卷积核。实验结果表明:文章的方法生成的卷积核的插值图在边缘上的最大波动率减小到1.54%。该方法更适用于包含纹理及文字等带丰富边缘特征的图像插值。
An edge optimized image interpolation convolution kernel search algorithm is proposed to tackle fake image fluctuation produced with Cubic or Lanczos interpolation convolution kernel.The interpolation convolution kernel is discretized by phase sampling,and the optimization loss with constraints is constructed.The experimental results show that the fluctuation rate on the edge of image inteplated with the proposed method is reduced to 1.54%.The method proposed in this paper is more suitable for image interpolation with rich edge features such as texture and text.
作者
梁骏
张明
LIANG Jun;ZHANG Ming(Hangzhou Nationalchip Technology Co.,Ltd,Hangzhou 310015,China;College of Information Science and Electronic Engineering,Zhejiang University,Hangzhou310058,China)
出处
《电视技术》
2019年第21期1-4,共4页
Video Engineering
关键词
图像插值
卷积核
边缘
最小值优化
image interpolation
convolution kernel
edge
minimum optimization
