Multi-scale error-driven dense residual network for image super-resolution reconstruction

Area of Science:

• Computer Vision
• Image Processing
• Deep Learning

Background:

• Current single-image super-resolution (SISR) methods struggle with multi-scale information and high-frequency detail extraction.
• Existing feed-forward network architectures lack iterative refinement and robust feedback mechanisms for enhanced high-frequency acquisition.

Purpose of the Study:

• To develop advanced strategies for multi-scale feature extraction, fusion, and feedback in SISR.
• To propose an innovative error-driven, multi-scale dense residual network (EMDN) for improved image super-resolution.

Main Methods:

• Utilized dual multi-scale features derived from convolutional kernels of varying sizes and diverse inputs, processed concurrently.
• Integrated an error-driven feedback mechanism into a dense residual network architecture.
• Developed an error-driven, multi-scale dense residual network (EMDN) for SISR.

Main Results:

• The proposed EMDN method outperformed existing approaches in both subjective and objective assessments across various scaling factors.
• Achieved improvements of up to 0.385% in peak signal-to-noise ratio (PSNR) and 0.191% in structural similarity index measure (SSIM) compared to baseline feed-forward networks.
• Demonstrated enhanced image resolution and restoration quality through comparative evaluations.

Conclusions:

• The EMDN effectively addresses limitations in multi-scale feature extraction and feedback mechanisms for SISR.
• Experimental results validate the effectiveness and practical significance of the proposed method for enhancing image resolution.
• The developed approach offers a robust solution for high-quality image reconstruction from low-resolution inputs.