一种基于GVF各向异性扩散模型的图像放大算法  

An Image Magnification Method with GVF-Based Anisotropic Diffusion Model

作　　者：李晓光[1] 沈兰荪[1] LAM Kin-Man 王素玉[1] 

机构地区：[1]北京工业大学信号与信息处理研究室,北京100124 [2]Department of Electronic and Information Engineering,The Hong Kong Polytechnic University,Hong Kong

出　　处：《电子学报》2008年第9期1755-1758,共4页Acta Electronica Sinica

基　　金：国家自然科学基金(No.60472036;60431020;60402036);教育部博士点基金(No.20040005015)

摘　　要：提出了一种基于梯度向量流各向异性扩散模型的图像放大方法.首先低分辨率图像插值放大作为高分辨率图像的初始估计,然后利用基于GVF的平均曲率扩散模型和高斯移动平均低分辨率模型约束进行迭代复原.GVF是一种有旋场,作为外力场用来描述图像的边缘特征,能够将初始图像中斜向边缘锯齿效应表示为流线型.采用GVF外力场约束平均曲率扩散过程,能够有效去除边缘锯齿现象并保持纹理结构.高斯移动平均模型提供了图像数据保真度约束,使结果更接近理想图像.实验结果表明,本文算法能够有效提高放大图像的主观视觉质量和客观PSNR.An image magnification method with GVF-based anisotropic diffusion model is proposed.An image is magnified by bilinear interpolation at first. Then, an iterative restoration with a GVF based mean curvature flow diffusion and a Gaussian moving average LR constraint is applied to the magnified image. Since GVF is a rotational field, as an external force field to descript the edges of an image, the vector flow will become streamline near the jagged edges. Therefore, the GVF based anisotropic diffusion will be helpful to remove the jagged effects as well as keep the texture structures. Meanwhile, the Gaussian moving average LR model provides a data fidelity constraint which makes the results more close to the ideal FIR images. Experiments results show that the proposed method can improve the quality of magnified image in terms of both the objective and subjective.

关 键 词：图像放大 梯度向量流 各向异性扩散 超分辨率复原 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]