基于自适应加权的正则化超分辨率图像重建  

Super resolution image reconstruction based on adaptive weighted regularization

作　　者：马飞[1] 苏敏[1] 杨晓梅[1] 

机构地区：[1]四川大学电气信息学院,四川成都610065

出　　处：《计算机工程与设计》2016年第5期1303-1308,共6页Computer Engineering and Design

摘　　要：传统正则化超分辨率方法中正则化参数是恒定值,在重建过程中不能根据图像局部平滑纹理特性自动调节,导致重建图像不能很好保留边缘纹理细节,为此提出一种基于L1范数的自适应加权全变分正则项和ADMM的超分辨率算法。根据目标图像局部结构信息自适应地为正则项加权,通过快速解耦算法将图像超分辨率转换为图像去噪,利用ADMM和快速傅里叶变换对正则化超分辨率模型进行有效的求最优解。多组对比仿真结果表明,与传统方法相比较,在视觉效果上,该算法更好地保留了重建图像的细节纹理,在峰值信噪比(PSNR)和结构相似度索引(SSIM)评价指标上有明显改善,提高了重建图像的质量。In the traditional method of regularized super-resolution,the parameter of regularization is,in the rebuilding process,a constant value,which can not adjust itself automatically according to local image smoothing texture features.The fixed parameter causes the edge and texture features of the reconstructed image can not be reserved well.A super-resolution algorithm based on adaptive weighted L1 norm total variation regularization and ADMM was proposed.The proposed method made the regularization weight automatically according to local structure of the target image information,and the super-resolution was transformed to an image denoising problem using the fast decoupled method.The optimal solution was got through the ADMM and fast Fourier transform.The experimental and visual results verify the superiority of the proposed method over the traditional method on preserving the texture features of the reconstructed image and improving peak signal-to-noise ratio（PSNR）and structural similarity（SSIM）.

关 键 词：超分辨率 图像重建 交替方向乘子法(ADMM) 自适应加权 全变分 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]