Data in the real world are usually high-dimensional and noisy, but in many cases, especially in computer vision research, we want to learn low-dimensional structures for problem processing. In order to solve the problems of duplicate data and noise in high-dimensional data, a video noise-reduction algorithm based on low-rank representation is proposed. By reducing the dimension of video data by principal component analysis (PCA), noise-reduction problem of video data is transformed into a low-rank matrix approximation problem. As a result, a low-rank matrix decomposition method is obtained. The resulting optimization problem is effectively solved by the augmented Lagrange multiplier (ALM). Experiments verify the effectiveness of the video noise-reduction method in the case of serious data destruction, and show that it is effective and robust compared with the existing methods.