基于鲁棒主成分分析的图像放大算法  

Image super-resolution based on robust principal component analysis

作　　者：范九伦[1] 张小丹[1] 徐健[1] 郭茹侠 

机构地区：[1]西安邮电大学通信与信息工程学院,陕西西安710121 [2]长安大学信息工程学院,陕西西安710064

出　　处：《西安邮电大学学报》2015年第6期37-44,共8页Journal of Xi’an University of Posts and Telecommunications

基　　金：国家自然科学基金资助项目(61340040;1202183;61102095)

摘　　要：给出一种采用鲁棒主成分分析去噪的图像超分辨率算法。对高分辨率训练图像进行Haar小波变换,使用鲁棒主成分分析法得到去噪后的近似子带字典和细节子带字典;将低分辨率测试图像的近似子带作为相应高分辨率测试图像的近似子带,通过细节子带字典恢复出高分辨率测试图像细节子带;通过逆Haar小波变换得到高分辨率测试图像,利用多级增强进一步提高图像的质量。实验结果显示,用所给方法得到的字典对噪声有鲁棒性,且高分辨率重建图像峰值信噪比较高。In order to improve the robustness of the low resolution images with noises, an image super- resolution algorithm based on PRCA denoising is proposed. Firstly~ the high resolution training images are transformed by Haar wavelet, and then the robust principal component analysis is employed to get denoised approximate subband dictionary and detail subband dictionaries. Secondly, the approximate subband of low resolution test image is regarded as its corresponding high resolution test image, and the detail subbands of high resolution test image are restored by detail subband dictionaries. Finally, the high resolution test image is reconstructed by inverse Haar wavelet transform. A multilevel enhancement process is used to further improve the image resolution. Experimental results show that the dictionary obtained by robust principal component analysis is robust to noises and singularities, and the high resolution image obtained by the proposed method has better visual effect and higher peak signal to noiseratio （PSNR）.

关 键 词：小波变换 字典学习 稀疏表示 鲁棒主成分分析法 超分辨率重建 

分 类 号：TN911.73[电子电信—通信与信息系统]