This paper proposes a novel anchor-based multi-view clustering method called CAMVC, which adaptively learns cluster-wise anchors to improve the discriminative representation in multi-view clustering. Unlike previous methods that focus on improving anchor diversity through orthogonal constraints, CAMVC introduces an anchor cluster assumption, where a prior cluster indicator matrix is used to guide the learning of cluster-wise anchors. This approach ensures that the learned anchors have a balanced and discriminative distribution, with within-cluster semantic consistency and between-cluster diversity. The method integrates latent anchor learning and consensus subspace representation construction into a unified optimization model. The key contributions include: (1) a novel anchor-based multi-view clustering method that adaptively learns discriminative anchors with cluster structure regularization and discriminative representation in a unified optimization framework; (2) an anchor cluster assumption introduced by defining a prior cluster indicator to learn diverse centroids of anchor clusters, ensuring within-cluster semantic consistency and between-cluster diversity; and (3) an alternating optimization algorithm to solve the proposed model. Extensive experiments on seven datasets demonstrate that CAMVC outperforms state-of-the-art methods in terms of clustering performance and efficiency. The method is efficient and scalable, with a time complexity of O(n), making it suitable for large-scale datasets. The results show that CAMVC achieves the best and second-best results in most cases, with the highest average scores and ranks across all metrics. The method is effective in learning discriminative subspace representations and is capable of handling large-scale multi-view clustering tasks.This paper proposes a novel anchor-based multi-view clustering method called CAMVC, which adaptively learns cluster-wise anchors to improve the discriminative representation in multi-view clustering. Unlike previous methods that focus on improving anchor diversity through orthogonal constraints, CAMVC introduces an anchor cluster assumption, where a prior cluster indicator matrix is used to guide the learning of cluster-wise anchors. This approach ensures that the learned anchors have a balanced and discriminative distribution, with within-cluster semantic consistency and between-cluster diversity. The method integrates latent anchor learning and consensus subspace representation construction into a unified optimization model. The key contributions include: (1) a novel anchor-based multi-view clustering method that adaptively learns discriminative anchors with cluster structure regularization and discriminative representation in a unified optimization framework; (2) an anchor cluster assumption introduced by defining a prior cluster indicator to learn diverse centroids of anchor clusters, ensuring within-cluster semantic consistency and between-cluster diversity; and (3) an alternating optimization algorithm to solve the proposed model. Extensive experiments on seven datasets demonstrate that CAMVC outperforms state-of-the-art methods in terms of clustering performance and efficiency. The method is efficient and scalable, with a time complexity of O(n), making it suitable for large-scale datasets. The results show that CAMVC achieves the best and second-best results in most cases, with the highest average scores and ranks across all metrics. The method is effective in learning discriminative subspace representations and is capable of handling large-scale multi-view clustering tasks.