Adversarial Multi-Path Residual Network for Image Super-Resolution.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Image Processing

Background:

• Deep convolutional neural networks (CNNs) have advanced single image super-resolution (SR).
• Existing SR methods often require deep, wide networks, leading to high computational costs and inefficiency.
• Current models struggle to simultaneously achieve high perceptual and objective quality in SR.

Purpose of the Study:

• To propose a novel, efficient network for single image super-resolution.
• To address the limitations of complexity, computational cost, and dual-quality assurance in existing SR methods.
• To develop a model that balances perceptual and objective quality with reduced parameters.

Main Methods:

• Introduced the Adversarial Multipath Residual Network (AMPRN).
• Developed a multi-path residual block (MPRB) for feature extraction with reduced parameters.
• Employed global gradual feature fusion and an adversarial gradient network with gradient loss.

Main Results:

• AMPRN significantly reduces network parameters compared to state-of-the-art methods.
• The proposed MPRB effectively extracts abundant local features.
• The adversarial gradient network improves gradient distribution, enhancing image quality.

Conclusions:

• AMPRN offers a more practical and efficient solution for single image super-resolution.
• The network achieves superior SR performance with a substantial reduction in parameters.
• AMPRN successfully enhances both perceptual and objective quality of super-resolved images.