In recent years, there has been increasing focus on utilizing multiple image modalities
to enhance various image processing tasks, including super-resolution(SR). One innovative
technique for generating a high-resolution (HR) image from its low-resolution (LR)
version employs another HR image modality for guidance. This method relies on joint
sparse representations derived from coupled dictionaries, capturing complex relationships
both similarities and differences between different image modalities in the sparse feature
space instead of the original image space. The overall framework is divided into two main
phases: the coupled dictionary learning phase and the coupled super-resolution phase.
During the learning phase, dictionaries are constructed using training data to link different
image modalities in the sparse feature domain. In the super-resolution phase, these
learned dictionaries are applied to reconstruct the HR version of the LR image, using the
related modality for guidance. An enhanced version of this technique incorporates a multistage
approach and neighborhood regression to achieve better performance. Extensive
testing on real multimodal image datasets shows that this method surpasses leading techniques,
reducing common issues such as texture inconsistencies between the guidance and
target images. Moreover, the model demonstrates greater resilience than deep learning
approaches, especially in noisy environments.