Soft-edge Assisted Network for Single Image Super-Resolution.

Area of Science:

• Computer Vision
• Image Processing
• Deep Learning

Background:

• Single Image Super-Resolution (SISR) is an ill-posed problem, challenging due to the difficulty of reconstructing high-frequency details.
• Existing Convolutional Neural Network (CNN)-based methods often increase network depth, leading to marginal gains and high computational costs.
• Integrating image prior knowledge offers a more efficient approach to image reconstruction.

Purpose of the Study:

• To propose a novel Soft-edge assisted Network (SeaNet) for high-quality Single Image Super-Resolution (SISR).
• To improve the efficiency and performance of super-resolution reconstruction by incorporating image soft-edge features.
• To address the limitations of conventional deep CNN approaches in SISR.

Main Methods:

• Developed SeaNet, a three-sub-network architecture comprising a Rough Image Reconstruction Network (RIRN), a Soft-Edge Reconstruction Network (Edge-Net), and an Image Refinement Network (IRN).
• Implemented a two-stage reconstruction process: Stage-I reconstructs rough SR features and soft-edges; Stage-II fuses these outputs for final high-quality SR image generation.
• Utilized soft-edge information as a crucial image prior to guide the reconstruction process.

Main Results:

• SeaNet demonstrates rapid convergence during training.
• The proposed method achieves excellent performance in high-quality SR image reconstruction.
• Experimental results validate the effectiveness of using soft-edge assistance in SeaNet for SISR.

Conclusions:

• SeaNet effectively reconstructs high-quality super-resolved images by integrating soft-edge priors.
• The proposed architecture offers an efficient and high-performing solution for Single Image Super-Resolution (SISR).
• Soft-edge information serves as a valuable prior for enhancing SR image reconstruction.