Deep Shearlet Residual Learning Network for Single Image Super-Resolution.

Area of Science:

• Computer Vision
• Image Processing
• Deep Learning

Background:

• Convolutional Neural Networks (CNNs) are used for single image super-resolution (SISR) by estimating residual images.
• Residual images contain high-frequency details and exhibit cartoon-like characteristics.

Purpose of the Study:

• To propose a Deep Shearlet Residual Learning Network (DSRLN) for estimating residual images in SISR.
• To leverage the optimal sparse approximation properties of shearlet transform for cartoon-like images.

Main Methods:

• The proposed DSRLN network is trained in the shearlet transform-domain.
• A dual-path training strategy and data weighting technique are employed to handle statistical variations in shearlet coefficients.

Main Results:

• The DSRLN scheme achieves comparable Peak Signal-to-Noise Ratio (PSNR) results to state-of-the-art deep learning methods.
• The proposed network requires significantly fewer parameters compared to existing methods.

Conclusions:

• The DSRLN is an effective approach for single image super-resolution, particularly for images with cartoon-like characteristics.
• The shearlet transform domain provides an efficient representation for residual image estimation in SISR.