张量低秩约束下的多帧图像去模糊

Multi-frame Deblurring Model Based on Low-rank Tensor

矩阵低秩估计模型在图像处理任务中有着广泛地运用。针对图像去模糊,利用矩阵低秩先验能保留图像的重要边缘信息从而实现去模糊。而对于多帧图像去模糊,基于矩阵的低秩模型并未充分考虑多帧图像间的时序和空间关系。针对该问题,我们提出基于三维张量低秩先验的多帧视频图像盲去模糊模型。在模型中,首先将多帧连续图像按时序维堆叠成张量,显式地考虑多帧图像间的时空关系,同时利用张量低秩先验约束保留图像重要纹理结构信息。利用交替迭代的方法求解模型,实现去模糊。通过在不同的数据集上实验结果表明,该方法能达到较好的去模糊结果。

Low-rank matrix approximation models have been successfully applied to numerous vision tasks.For image deblurring,the use of low-rank prior can retain the important edge information.However,for multi-frame image deblurring,the low-rank matrix models do not consider the temporal and spatial relationship among multi-frame images.In this paper,we propose a low-rank prior deblurring model based on a three-dimensional tensor,we pile up the continuous images into a tensor according to the temporal dimension,and use the tensor low-rank prior constraint,which not only maintains the dominant texture structure information to estimate the blur kernel but also considers the spatiotemporal relationship among multiple frames.We solve the energy model by alternating direction minimization method.The experimental results on different datasets show that our method can achieve better results.