针对高维度矩阵的低秩恢复问题中核范数与l1范数过惩罚导致的结果偏差,提出一种矩阵恢复方法.使用准范数代替低秩恢复问题中常用的核范数约束,使用零范数代替l1范数约束.对于准范数的求解问题,采用与准范数等价的Fro-benius/核混合范数进行替代,提出基于交替近似的线性最小化方法对目标函数进行求解.在合成数据与真实数据上的实验结果表明,该算法在主观视觉效果与客观数值比较上都能获得更好的结果.

为解决因运动模糊导致的图像关键信息丢失问题,提出一种高效盲去模糊算法.鉴于图像的模糊过程会显著降低图像在频域的表达,同时也会显著改变清晰图像的稀疏性,提出基于图像卷积谱和L0正则化的鲁棒去模糊算法,分别对图像与模糊核进行约束.该算法基于分离变量的方法,保证每个子问题都有良好的收敛性,在文本与自然图像去模糊任务中都取得了良好的效果.

在视频预测领域,传统的CNN与LSTM都不能充分表征视频中的时空特征.针对这一问题提出空间自适应卷积LSTM算法.受空间变换网络启发,在卷积LSTM内部的"input-to-state"计算过程中将传统卷积操作改为空间自适应卷积:利用额外卷积层获得自适应卷积所需的位置参数,令自适应卷积根据时空信息选择卷积位置,提升模型捕捉时空变换特征的性能;并针对雷达回波预测提出多分支编码预测的网络架构,根据降水类别训练4个不同的支路,以提升网络的预测性能.在合成数据集与真实数据集上的实验结果表明,该模型取得了有竞争力的结果,单独设计一个模块让网络显式地学习某种特征会使网络有更好的性能.

Blind motion deblurring is one of the most basic and challenging problems in image processing and computer vision. It aims to recover a sharp image from its blurred version knowing nothing about the blurring process. Many existing methods use the maximum a posteriori or expectation maximization framework to tackle this problem, but they cannot handle well the natural images with high-frequency features. Most recently, deep neural networks have been emerging as a powerful tool for image deblurring. In this paper, we show that encoder–decoder architecture gives better results for image deblurring tasks. In addition, we propose a novel end-to-end learning model that refines the generative adversarial network by many novel strategies to tackle the problem of image deblurring. Experimental results show that our model can capture high-frequency features well, and achieve the competitive performance.