<p>Incomplete multi-view clustering is a significant and challenging task in machine learning and data mining. However, current methods still face the following issues: First, most existing approaches fail to fully utilize complementary information between views to recover missing data, limiting the potential for improved clustering performance. Second, the prevalence of missing views presents substantial challenges for current methods in capturing the complex interactions between views, resulting in inaccurate and unstable clustering outcomes. To address these issues, this paper proposes a novel framework: Dual Recovery and Multi-scale Feature Enhancement (DRMFE). Specifically, the framework employs an attention-based encoder to extract features and utilizes a dual recovery mechanism combining self-reconstruction and cross-view prediction to effectively recover missing view data in the latent space. Additionally, the proposed multi-scale feature enhancement module optimizes data representation in the latent space by modeling local similarity and preserving global structure, thus focusing on both local details and global trends. Finally, during the feature fusion stage, we employ feature synergy consistency optimization to mitigate the noise introduced by the fusion of unaligned features. Experimental results show that DRMFE outperforms existing state-of-the-art methods on multiple benchmark datasets.</p>