基于放大曲波基的方向性超分辨率图像重构技术

Directional Reconstruction of Super Resolution Image by Magnifying Curvelet Basis

为解决传统图像放大算法边界视觉效果不佳的问题,提出基于二代曲波变换的方向性超分辨率图像重构算法.对图像进行j层曲波分解,利用不同尺度上曲波基的空间比例关系获得放大图像j层分解系数,通过最外层曲波基空间模型可构建(j+1)层放大图像的曲波分解系数,采用新的非线性函数对全部曲波系数进行增强处理,根据曲波分解的方向性,最终可通过曲波重构获得边缘特征较好的放大图像.实验结果表明,基于曲波方向性图像放大算法,可以较好地保留原图的几何特征,增强边缘清晰度;将两幅典型图像放大后的峰值信噪比与经典方法(差值算法)比较分别提升了2.2及0.6 dB.

A super-resolution image reconstruction algorithm was proposed using the 2nd generation curvelet to reduce the edge blur caused by traditional algorithms.In the proposed algorithm,the original image is decomposed into j scales using curvelet.The curvelet coefficients in the j scales of the zoomed-in image are obtained by utilizing the proportionality of curvelet bases between adjacent scales,and the curvelet coefficients in the（j＋1）th scale are determined by utilizing the spatial template of curvelet coefficients with the largest scale number.All the curvelet coefficients are processed with a nonlinear function to enhance image quality.The zoomed-in image with fine edges is finally created through curvelet reconstruction because of the good directional characteristic of curvelet.Experiments on two benchmarking images shown that,the proposed algorithm could preserve more image features and edge sharpness,and the peak signal to noise ratios（PSNRs） for the two images increased by 2.2 and 0.6 dB,respectively,compared with those obtained with a traditional interpolation algorithm.