基于整体最小二乘法的高分辨率图像重建算法

Total least squares based high-resolution image reconstruction algorithm

提出一种用于高分辨率图像重建的整体最小二乘算法。在现有多数重建算法中,假设系统矩阵是精确的而误差主要源于采样图像,但实际上抖动误差也出现在系统矩阵中。该方法能同时最小化这两种误差,采用基于正则化的Rayleigh商来光滑解,用共轭梯度算法来迭代求解该正则化Rayleigh商的最小化函数。实验证明该方法对于抖动系统矩阵是稳定和精确的。

The paper presents a High-Resolution（HR） image reconstruction algorithm based on total least squares approach.Many HR algorithms assume that system matrix is accurate and errors are from the sampled Low-Resolution（LR） images.In practice,errors also occur in the system matrix.The paper proposes the TLS approach that minimizes the two kinds of errors.A Rayleigh Quotient（RQ） formulation based on regularization is used to enforce a smooth solution.Then a conjugate gradient algorithm is used to minimize the modified RQ function.Experiment results show that the method is stable and accurate for the perturbation system matrix.