在很多实际应用中很难获得适当的图像训练集，但是单幅图像放大却是一个病态问题。利用图像局部结构的自相似性和可传递性，结合非下采样Contourlet变换(NSCT)的优点，提出一种基于自类推与NSCT的单幅图像超分辨率重建(SRR)方法。采用NSCT对源图像和退化图像进行多尺度、多方向分解，得到用于学习的各带通方向子带对，利用图像自类推技术生成高分辨率的各带通方向子带，与立方插值放大后的源图像进行NSCT重构得到超分辨率重建图像。实验结果表明，该方法可以独立进行，摆脱一般方法对训练结合的依赖，能产生更为合理的细节，视觉边缘更清晰，图像更逼真。

In lots of cases, it′s difficult to obtain appropriate image training set, but single-image zooming is an ill-posed problem. Using the self-similarity feature among local structure in an image which can be maintained in the scale space and the advantage of nonsubsampled contourlet transform (NSCT), a single image super-resolution reconstruction (SRR) algorithm based on image analogies and NSCT is proposed. NSCT is performed on the original image and the degraded image at different scales and directions, thus varieties of directional bandpass subband pairs are obtained. The relationships between the subband pairs by image self-analogies are learned to generate high resolution varieties of directional bandpass subband. The super-resolution reconstructed image is obtained by transforming these changed subband coefficients and the zoomed-original image by bicubic interpolation back to the spatial domain. The experimental results show that the algorithm can be executed independently without any supposed outliers. It can generate more reasonable details than general methods, thus the edges are much clearer and the image is more natural-looking.