Incomplete multi-view clustering (IMC) is crucial for real-world applications such as image retrieval, text classification, and subtype recognition, where multi-modal data frequently suffer from missing values and severe noise due to acquisition limitations or transmission losses. However, most existing methods rely on raw data completion, single-tensor constraints, and isolated multi-step processing paradigms. These approaches not only struggle to simultaneously preserve cross-view consistent principal components and filter out view-specific variations, but their phased nature also introduces irreversible error accumulation. To overcome these bottlenecks, this paper proposes a novel unified framework named Dual-Tensor Graph Completion for Incomplete Multi-View Clustering (DTGC-IMVC). Specifically, to eliminate reconstruction noise at the data feature level, an adaptive similarity graph completion mechanism is introduced to directly recover the similarity graph within the self-representation. Meanwhile, a dual-tensor architecture seamlessly integrating the tensor nuclear norm and the Schatten-p norm is constructed to rigorously capture high-order cross-view consistent features and effectively sparsify noise and view-specific components. Furthermore, a unified multi-source denoising strategy constrained by the L2,1-norm is introduced. Comprehensive experiments on various benchmark datasets demonstrate that the proposed framework substantially outperforms state-of-the-art methods in terms of clustering performance and robustness. The source code of this work is available at https://github.com/kamenrider-x/DTGC-IMVC.