Reconfigurable Dimensionality Reduction Based on Joint Dictionary Learning and Applications

In recent years, sparse representation has been widely used in many fields, such as image denoising, facial recognition, image classification, etc. However, some traditional sparse representation algorithms have weak discriminative ability because the model results are too simple and cannot learn a specific dictionary for images. In this paper, we propose a method to learn recoverable low-dimensional representations based on sparse representations for large-scale datasets and to solve the pairwise dictionary (D, P) of the learning problem in the dimensionality reduction process by introducing Johnson-Lindenstrauss Lemma. We use a single super-resolution image and facial image for experiments, which show that our proposed pairwise dictionary learning model has a stronger performance in compression and reconstruction than other reconstruction models. This paper demonstrates that learning and discriminating in the compression domain is feasible.