基于矢量量化的模糊参数辨识及分辨率增强方法

在实际的图像复原中通常需要预先估计模糊函数。该文将Nakagaki等人提出的基于矢量量化的模糊参数辨识算法进行改进，利用Sobel算子形成特征矢量，避免了LOG滤波器参数的选择，增强了算法对辨识不同类型图像的模糊函数的鲁棒性，并利用DCT对特征矢量降维，减小了计算量。同时将其应用于超分辨率图像复原中，辨识出多幅低分辨率图像的模糊函数，然后融合具有不同模糊函数和信噪比的低分辨率图像，实现了盲超分辨率图像复原。仿真结果表明了改进算法的有效性和可行性。

A VQ-Based Parameter Identification Approach to Blind Image Restoration and Resolution Enhancement

Blur identification is usually necessary in image recovery. In this paper, an improved approach is proposed on the basis of VQ-based blur identification algorithm developed by Nakagaki, In this method, Sobel operator is used for extracting feature vectors, so that the selection of the parameter of the LOG filter is avoided and this method is robust to different types of images. The dimensionality of the vector is reduced by utilizing DCT. Meantime, extension of this method to blind super-resolution image restoration is achieved. After blur identification, a super-resolution image is reconstructed from several low-resolution images obtained by different loci. Simulation results demonstrate the feasibility and validity of the method.