<p>Multi-view clustering can obtain a comprehensive clustering result by integrating the data information across multiple views. Based on this fact, some scholars have developed numerous multi-view clustering algorithms that have demonstrated excellent clustering performance. However, these algorithms may still possess certain limitations. For example, the data with noise as a dictionary can pose a multitude of research challenges in self-representation theory. To address the above issue, we propose a new multi-view clustering algorithm based on low-dimensional structure and global representation (MVCLG). Specifically, we use a self-representation model to extract the global consistent information of all-view data. In the self-representation model, the data as a dictionary may have noise. Hence we use non-negative matrix factorization (NMF) to uncover the low-dimensional structure of each-view data and remove some redundant information related to noise. In this way, the clean low-dimensional structures can better guide the global representation so that a high-quality clustering result is obtained. In addition, an adaptive loss function and automatic weighting strategy are implemented in our model. Finally, we design the iterative update optimization strategy and prove its convergence. Compared with ten state-of-the-art algorithms, experimental results on seven real-world datasets demonstrate that our proposed algorithm achieves superior effectiveness. The code of MVCLG is published at the following website <a href="https://github.com/xzwxing/MVCLG">https://github.com/xzwxing/MVCLG</a>.</p>