<p>With the widespread application of deep learning in computer vision, effectively handling large volumes of unlabeled data has become a significant challenge. Traditional unsupervised visual clustering methods often face issues such as high computational complexity and substantial resource consumption, particularly on large-scale datasets, where they struggle to meet efficiency requirements. Quantum computing, with its parallelism and quantum superposition properties, offers potential advantages for processing large-scale data. However, existing quantum computing-based visual clustering methods often fail to fully leverage the efficient modeling and feature interaction capabilities of quantum information, resulting in limitations in feature representation and relationship modeling. To address these issues, this paper proposes a Quantum Self-Attention Clustering (QSAC) method for unsupervised visual clustering. QSAC integrates a quantum self-attention mechanism with classical neural networks, utilizing quantum rotation gates to generate quantum states for queries (Q), keys (K), and values (V). By designing global entanglement gates, QSAC introduces entanglement relationships among qubits, enabling feature aggregation and global dependency modeling to capture complex patterns and relationships in the data. Unlike classical methods, QSAC encodes features into quantum superposition states, compressing high-dimensional image features into the state space of qubits, thereby reducing feature dimensionality. The parallelism of quantum superposition enhances the efficiency of this process, avoiding the computational resource bottlenecks of traditional methods. Experimental results demonstrate that QSAC outperforms classical clustering methods on multiple benchmark datasets, not only reducing computational resource demands but also significantly improving clustering accuracy.</p>