Object-Level Matching for Multi-source Image Using Improved Dictionary Learning Algorithm

Solving the problem of multi-source information matching is the foundation of multi-source information fusion. Aiming at the heterogeneity between different multi-source images, we propose a new method of object-level matching for multi-source image based on improved dictionary learning. Two main steps, unified representation and similarity measure, are contained. Firstly, we complete the unified representation of multi-source images by improved dictionary learning algorithm. On the basis of the representation ability of dictionary learning, we further make full use of the label information to boost the discriminative ability of dictionary, which is beneficial to the implementation of object matching. Then, we construct a neural network to learn the distance metric standard between matching and non-matching by supervised learning, which can replace the traditional distance metric method. In addition, we produce two sets of multi-source image object matching datasets based on the open datasets, which verifies the validity and accuracy of the algorithm, and shows the good performance of the algorithm in solving the zero-shot learning problems.