ABSTRACT
Image feature matching is a critical task in computer vision. Traditionally, feature matching is processed independently for each feature. However, matched features can provide constraints for subsequent matches, thus improving overall matching efficiency. In this paper, we propose employing the vector field consensus (VFC) model to restrict subsequent feature matching. We incorporate the concept of the Gaussian process with VFC to estimate the covariance of the vector field of a new feature. This covariance allows us to determine the search range of this new feature in the second image. As the VFC model is progressively updated, the search range is further reduced, making the matching process more efficient. To evaluate our proposed framework, we conducted an experiment. The results demonstrate that our framework is more efficient than the traditional nearest-neighbor approach and the original vector field consensus algorithm.
