Image superresolution reconstruction via granular computing clustering

Area of Science:

• Computer Vision
• Artificial Intelligence
• Image Processing

Background:

• Single image super-resolution (SR) is a challenging task in computer vision.
• Existing methods like bicubic interpolation, sparse representation, and NNLasso have limitations in reconstructing high-fidelity SR images.

Purpose of the Study:

• To propose a novel approach for single image super-resolution using granular computing (GrC) clustering.
• To improve the accuracy and quality of super-resolved images from low-resolution inputs.

Main Methods:

• Training images are partitioned into super-resolution (SR) and low-resolution (LR) patches.
• Granular computing (GrC) clustering is employed, utilizing hypersphere representation and fuzzy inclusion measures.
• A granule set (GS) is induced by GrC to establish a relationship between LR and SR images via lasso regression.

Main Results:

• The proposed GrC clustering method achieved the lowest root mean square errors (RMSE) compared to bicubic interpolation, sparse representation, and NNLasso.
• Experimental results demonstrate superior performance in reconstructing SR images.

Conclusions:

• Granular computing clustering offers a promising and effective method for single image super-resolution.
• The approach significantly reduces reconstruction errors, leading to higher quality SR images.