Image clustering is a hot research topic at present, and nonnegative matrix factorization (NMF) has been widely used in the field of image clustering. However, a single NMF clustering algorithm can’t be applied to all datasets, and the NMF algorithm directly processes the original space of the data, which has poor noise resistance. Ensemble clustering can solve the above problems. Ensemble clustering combines several basic clustering results into a consistent result, which not only improves the quality of clustering, but also enhances the robustness of the algorithm. Therefore, a hierarchical preprocessing NMF weighted integrated clustering algorithm is presented. The algorithm introduces the idea of hierarchical division, ensemble clustering and bipartite graph into the NMF algorithm. In the preprocessing stage, the number of clusters is obtained by hierarchical division. The co-association matrix is then obtained by local weighting. Finally, the final clustering result is obtained by partitioning using the consistency function based on the bipartite graph. The algorithm is tested on five datasets to verify the effectiveness of the algorithm over traditional algorithms and other ensemble algorithms.