The super-resolution reconstruction algorithm of multi-scale dilated convolution residual network.

Area of Science:

• Computer Vision
• Deep Learning
• Image Processing

Background:

• Traditional super-resolution algorithms struggle with small receptive fields and feature information loss.
• Extracting multi-scale features is crucial for effective image reconstruction.

Purpose of the Study:

• To propose a novel super-resolution reconstruction algorithm using a multi-scale dilated convolution network.
• To address limitations of existing methods by enhancing feature extraction and fusion.

Main Methods:

• Utilizes dilated convolution kernels with varying receptive fields for multi-scale feature extraction.
• Employs residual attention dense blocks and local residual connections for feature fusion.
• Incorporates residual nested networks and jump connections to accelerate convergence and prevent degradation.

Main Results:

• The proposed algorithm demonstrates superior performance on standard datasets (Set5, Set14, BSDS100, Urban100).
• Achieved higher peak signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM) compared to established algorithms.
• Reconstructed images exhibit improved visual quality.

Conclusions:

• The multi-scale dilated convolution network effectively enhances super-resolution reconstruction.
• The proposed method offers a significant improvement over existing super-resolution techniques.
• The algorithm's ability to fuse multi-scale features and nonlinear expressions boosts reconstruction performance.