摘要
稀疏表示法在单幅图像超分辨率重建问题中受到广泛的关注。本文介绍了一种使用稀疏表示进行超分辨率图像重建的方案。该方案首先由低分辨率的输入图像块求取稀疏表示系数,然后根据此系数生成对应的高分辨率图像块,最后由高分辨率块重建出整幅图像。在求取稀疏表示系数时,本文采用了一种借助预处理共轭梯度算法计算搜索方向的内点方法。仿真结果表明,本方案在视觉感受和客观量化两方面都比现有的双三次插值法和最小角度回归(LARS)法具有更好的性能。与双三次插值法和LARS法相比,本方法所得图像的均方根误差(RMSE)值通常可以分别改善0.29和0.7。
A lot of attention has been paid to sparse representation in single image super-resolution. A scheme is employed to implement the image super-resolution via sparse representation. First, the sparse representation from low-resolution / input patches is sought by sparse representation algorithms. Then the corresponding high-resolution outputs are generated from them. Finally, the whole high-resolution image from patches is reconstructed. To get the sparse coefficients, an Interior Point Method （1PM） is adopted, which uses preconditioned conjugate gradients algorithm to compute the search direction. Simulation results show that our scheme outperforms the existing bicubic interpolation, Least Angle Regression （LARS） and other algorithms in both visually and qualitative evaluations. Typical Root-Mean-Square Error （RMSE） reduction of 0.29 and 0.7 is achieved over Bicubic and LARS algorithms, respectively.
出处
《光电工程》
CAS
CSCD
北大核心
2012年第6期125-130,共6页
Opto-Electronic Engineering
基金
国家自然科学基金(61075013)
中国博士后科学基金资助项目(20100471671)
关键词
稀疏表示
图像超分辨率重建
内点法
预处理共轭梯度法
sparse representation
image super-resolution reconstruction
interior point method
preconditioned conjugate gradients
