SUPERVISED COUPLED DICTIONARY LEARNING FOR MULTI-FOCUS IMAGE FUSION

Abstract

Among all methods that have tackled the multi-focus image fusion problem, where a set of multi-focus input images are fused into a single all-in-focus image, the sparse representation based fusion methods are proved to be the most effective. Majority of these methods approximate the input images over a single dictionary representing only the focused feature space. However, ignoring the blurred features sets limits on the sparsity of the obtained sparse representations and decreases the precision of the fusion.

This work proposes a novel sparsity based fusion method that utilizes a joint pair of dictionaries, representing the focused and blurred features, for the sparse approximation of source images. In our method, more compact sparse representations (obtained by using both features in the sparse approximation), and classification tools (provided by using the two known subspaces (focused and blurred)) are exploited to improve the performance of the existing state of the art fusion methods. In order to achieve the benefits of using a joint pair of dictionaries, a coupled dictionary learning algorithm is developed. It enforces a common sparse representation during the simultaneous learning of two dictionaries, fulfils the correlation between them, and improves the fusion performance. The detailed comparison with the state of the art fusion methods shows the higher efficiency and effectiveness of the proposed method.