Graph collaborative filtering stands as a pivotal technology within recommendation systems, showcasing its significance and efficacy across diverse application scenarios. This approach delineates users (items) by explicitly capturing neighboring information to discern user preferences for items. Despite its effectiveness, collaborative filtering-based recommendation methods encounter challenges due to data sparsity. Recent approaches have turned to contrastive learning to mitigate these challenges, albeit solely constructing augmented views through random perturbation methods. However, reliance solely on contrastive learning proves inadequate. It overlooks the interrelations among high-order user (item) neighbors, thus failing to harness their information fully. Even when some methods consider higher-order neighbors, they merely acquire their information without effectively filtering out noise. This paper introduces HGACF, a novel homogeneous neighbor graph contrastive learning method designed for collaborative filtering. HGACF strategically leverages higher-order neighbor information. Specifically, we construct two homogeneous neighbor graphs derived from the user-item graph, forming neighbor nodes of the same type. This graph facilitates the extraction of high-order neighbor relationships among users (items), subsequently employing various aggregation strategies across different graphs to consolidate valid information. Furthermore, we propose a comparison learning paradigm between graphs, creating comparison tasks for optimization purposes. Extensive experiments on four public datasets demonstrate the effectiveness of the proposed HGACF.