基于协方差特征的图像内插方法的研究

本文提出了一种改进图像插值的方法,该方法通过提取低分辨率图像的协方差矩阵和协方差向量估算出高分辨率图像的相应参数,然后经过内插得到高分辨率图像。其突出优点是它的边缘自适应特性。经过实验证明,用这种方法获得的高分辨率图像和其他方法相比,图像的峰值信噪比(ＰＳＮＲ)有显著提高。在图像边缘模糊和阶梯形失真不仅比传统的线性内插方法有所降低,即使与基于凸面投影(ＰＯＣＳ)等方法相比图像也更加清晰。同时由于算法避免了复杂的迭代计算,因而运算的速度更快。

Covariance Feature Based Image Interpolation

In this paper, a fast and edge preserved image interpolation method is proposed. Using the covariance matrix and covariance vector in low resolution image, it is possible to estimate the corresponding information in high resolution image, thus obtains each high resolution pixel values. The most attractive advantage of this method is its edge orientation adaptive property. Using this method we can dramatically raise PSNR over the other methods. The edge blurring and ringing artifacts have been relatively reduced in compare to linear interpolation. Moreover, the method shows its advantage over POCS based interpolation. And at the same time, without using time consuming iterative algorithm, its computation complexity is also moderate.