<p>Video super-resolution algorithms have a wide range of applications, but most traditional super-resolution algorithms are designed based on vector-variable models, which have relatively high computational complexity. Most super-resolution algorithms based on deep learning are supervised. To reduce the complexity of traditional vector-variable models and leverage the unsupervised nature of traditional algorithms, this paper proposes an iterative optimization super-resolution algorithm based on matrix-variables. Firstly, a matrix super-resolution degradation model with constraints is introduced. Secondly, introducing TV regularization term and residual regularization term, by utilizing inter-frame residual information to improve reconstruction accuracy while maintaining the stability of video reconstruction in the timeline direction. Finally, the discrete recurrent neural network model was extended to iteratively solve the proposed optimization problems. Compared to traditional multi-channel super-resolution algorithms, the practical improvement is about 70&#xa0;s, and the PSNR has increased by 0.5dB. Compared with depth based super-resolution algorithms, PSNR has been improved by 0.9dB. So the experimental results confirm that the proposed algorithm has advantages in terms of running time and reconstruction accuracy.</p>