Single Image Superresolution via Directional Group Sparsity and Directional Features.

Area of Science:

• Computer Vision
• Image Processing
• Signal Processing

Background:

• Single image super-resolution (SR) reconstructs high-resolution (HR) images from low-resolution (LR) inputs.
• Reconstruction is challenging due to missing details in LR images, necessitating effective prior knowledge.

Purpose of the Study:

• To propose a novel SR method exploiting directional group sparsity and directional features for improved reconstruction.
• To enhance image quality by addressing missing details in single image super-resolution.

Main Methods:

• Utilized curvelet transform for extracting directional features for region selection and weight estimation.
• Developed a combined total variation regularizer assuming group sparsity of natural image gradients.
• Incorporated a directional nonlocal means regularization term considering pixel values and directional information to reduce artifacts.

Main Results:

• The proposed method demonstrated superior performance in quantitative metrics like Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity (SSIM).
• Qualitative results confirmed higher quality SR reconstruction compared to state-of-the-art algorithms.
• The energy function was minimized using a framework for first-order conic solvers.

Conclusions:

• The novel SR approach effectively exploits directional group sparsity and features for superior reconstruction.
• The method successfully suppresses artifacts and achieves high-quality results in single image super-resolution.
• This work advances the field of image super-resolution through innovative regularization techniques.