带稀疏惩罚的LCNN盲目图像复原

为了加强算法的稀疏性和稳定性,在SCAD基础上提出了一种新的稀疏惩罚函数,并加入到拉格朗日约束神经网络中,以克服传统盲源分离方法和独立分量分析方法的缺陷,有效地避免了方程的病态问题,提高盲目图像复原的稀疏性、稳定性和准确性。通过人工数据和真实数据的不同复原算法对比实验,证明了带稀疏惩罚的拉格朗日约束神经网络盲目图像复原技术具有良好的图像复原效果。

Blind Image Restoration Using LCNN with Sparse Penalty

In order to improve sparsity and robustness, a novel sparse penalty function based on smoothly clipped absolute diviation (SCAD) is proposed and applied to Lagrange Constraint Neural Network (LCNN). This method can solve ill-conditioned problem and improve sparsity, stability, and accuracy in blind image restoration. Both artificial and real-world data are calculated under some different restoration methods. Results of the experiments show that Lagrange constraint neural network with sparse penalty has better restoration effect.