<p>Image matching is a fundamental task in computer vision, widely applied in image registration, fusion, and retrieval. Image matching can be seen as a similarity calculation challenge for patches. Visible and infrared images exhibit distinct imaging principles, leading to significant dissimilarities between the two modalities. The disparity in features between visible and infrared images poses difficulties in measuring similarity, thereby impeding effective visible-infrared image matching. In this paper, we propose a visible-infrared image matching network called CMatchnet. We design a network Cnet for feature extraction. We use GELU and Group Normalization in CNet to improve network performance while reducing network parameters, and use the concept of "cardinality" to enhance the network’s feature extraction capabilities. CNet demonstrates proficient feature extraction capabilities for both visible images and infrared images with great differences. To achieve accurate matching, we employ a multiple adaptive prediction strategy. The multiple adaptive prediction strategy can get abundant matching information. Furthermore, we curate and annotate an existing visible-infrared image dataset to ensure ample training and testing data. The experimental results demonstrate the superiority of the proposed method.</p>