Feature Matching with Similarity Domains Network

Feature matching is an essential step in many computer vision applications. Similarity Domains Network (SDN) is a type of Neural Networks and has been recently proposed. SDN computes multiple kernel parameters to define the decision function. In this paper, we describe how to utilize the computed kernel parameters of SDN for feature matching. Normally, using only the distance value between the closest features does not yield good matches. However, in our experiments, we demonstrate that we can match features by using the distance value between the closest Gaussian kernel centers (also known as similarity domain centers) that are computed by SDN.