SeaNet

SeaNet reconstructs high-resolution images from single low-resolution inputs by integrating soft-edge priors into a multi-network architecture for single image super-resolution.

Key Features:

• Architecture Composition: SeaNet comprises three sub-networks: Rough Image Reconstruction Network (RIRN), Soft-edge Reconstruction Network (Edge-Net), and Image Refinement Network (IRN).
• Rough Image Reconstruction Network (RIRN): Generates initial rough super-resolution feature maps from the low-resolution input.
• Soft-edge Reconstruction Network (Edge-Net): Reconstructs soft-edge features to capture edge details and high-frequency information.
• Image Refinement Network (IRN): Fuses outputs from RIRN and Edge-Net and refines them to produce the final high-resolution image.
• Two-stage Reconstruction Process: Stage-I runs RIRN and Edge-Net concurrently to produce rough SR maps and soft-edge features, and Stage-II fuses those outputs for IRN-based refinement.
• Soft-edge prior integration: Explicitly leverages soft-edge information as an image prior to guide reconstruction of high-frequency details.
• Efficiency and Performance: Integration of soft-edge information enables rapid convergence and reduces the need for excessively deep networks, improving computational efficiency and output quality.

Scientific Applications:

• Medical imaging: Enhances resolution of low-resolution medical images to improve visualization of anatomical or pathological details.
• Satellite imagery: Reconstructs higher-resolution satellite images from low-resolution acquisitions for improved remote sensing analysis.
• Digital forensics: Restores fine details in low-resolution forensic images to support identification and evidentiary analysis.

Methodology:

SeaNet employs three explicitly defined sub-networks (RIRN, Edge-Net, IRN) in a two-stage process where Stage-I concurrently reconstructs rough SR feature maps and soft-edge features and Stage-II fuses those outputs for IRN-based refinement.