Simultaneous High-Resolution Blind Image Reconstruction and Perturbation Defense
High-resolution image reconstruction from partial observations plays an important role in modern electronic information systems and has attracted widespread intense attention in both the academic and industrial fields. It is often difficult for the traditional handcrafted prior image reconstruction...
Saved in:
Published in: | IEEE access Vol. 12; pp. 36173 - 36182 |
---|---|
Main Authors: | , , , , , , , |
Format: | Journal Article |
Language: | English |
Published: |
Piscataway
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
2024
IEEE |
Subjects: | |
Online Access: | Get full text |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Summary: | High-resolution image reconstruction from partial observations plays an important role in modern electronic information systems and has attracted widespread intense attention in both the academic and industrial fields. It is often difficult for the traditional handcrafted prior image reconstruction methods to recover delicate image details because of the inferior prior characterization abilities of these methods; this is especially true in the presence of complex electromagnetic perturbations in real-world environment scenes. Therefore, in this paper, an efficient image high-resolution reconstruction and perturbation defense model, named DGMR, is proposed based on deep Gaussian mixture learning. In particular, a simultaneous maximum a posterior (MAP) reconstruction-defense framework is designed based on the learned deep Gaussian mixture prior. Moreover, a channel attention mechanism is designed for image spatial correlation exploitation. Both the external and internal information are explored using a deep Gaussian mixture. A deep residual Swin transformer module is constructed to further characterize the image and learn Gaussian mixture priors, including both the image means and variances; in contrast, existing methods calculate only the image means but ignore the variances. Furthermore, sparse regularized united learning is developed to improve invariant representation learning ability, and the model is customized by internal learning with spatial constraints and regularization. Extensive qualitative and quantitative experiments are performed, confirming that DGMR is superior to the existing state-of-the-art systems. |
---|---|
ISSN: | 2169-3536 2169-3536 |
DOI: | 10.1109/ACCESS.2024.3370165 |