Abstract:

This paper proposed an interpolation algorithm using wavelet transform based on statistical features for natural images. The basic idea was to estimate the high-resolution covariances from low-resolution counterpart based on their "geometric duality" so that un-sampled pixels were obtained. In addition, wavelet decomposition and reconstruction were also used to enhance image details. No iterative computation was needed in this algorithm. The application of this new interpolation algorithm shows that it can keep the contours and details of non-edged areas successfully.

Key words:  image interpolation, wavelet transform, statistical feature

摘要：

针对一般图像插值方法常见的两个缺点:边缘锯齿、细节模糊,提出了一种基于图像统计特征的小波变换插值方法,该方法利用图像插值前后边缘统计特征的相似性,从低分辨率图像中获取估算高分辨率图像所需的协方差矩阵和向量,保持了边缘光滑性。同时利用小波分解与重构,将图像非边缘的细节特征进行了增强。该方法无需迭代运算,计算量较小。实验结果表明了用该方法获取高分辨率图像的有效性和可行性。

关键词: 图像插值, 小波变换, 统计特征