A local feature descriptor for image matching with CNN and Transformer

Feature matching is critical in computer vision and has the great variety of applications in diverse fields, such as image fusion, image retrieval, target tracking, and 3D reconstruction. At present, there are various of image acquisition, and the images acquired by different sensors have different perceptual information. In contrast, the current method of feature matching matches few feature points, which is incapable of effectively synthesizing and exploring the feature information of different images. We designed a composite CNN and Transformer structure to address this issue. We then performed experiments on publicly available datasets to test the model's capabilities. It has been shown experimentally that our model has a strong match effect.