基于即插即用框架的脉冲噪声图像复原算法

Image restoration algorithm based on plug-and-play regularization to remove impulse noise

针对脉冲噪声下的图像复原存在全变分正则化方法会在图像的光滑区域产生阶梯效应、L1保真项不能很好地刻画真实图像与退化图像的关系等问题,本文结合卷积神经网络和Lp保真项,提出一种即插即用的数学优化模型,利用交替方向迭代算法求解所提出的模型。数值实验结果表明,与L1保真项相比,所提出模型获得的复原图像能够有效地去除脉冲噪声,并在边缘、纹理和光滑度等方面有较好的应用效果。

For the staircase effect in smooth regions of the image and the inability of L1 fidelity terms to accurately depict the relationship between real and degraded images,this paper proposes a plug and play mathematical optimization model by using alternating direction iteration algorithm that combines convolutional neural networks and Lp fidelity terms to address the problems of total variational regularization methods in image restoration under pulse noise.The numerical experimental results show that compared with the L1 fidelity term,the restored images obtained by the proposed model can effectively remove pulse noise and maintain good edge,texture,smoothness and other aspects.