Regional spatially adaptive total variation super-resolution with spatial information filtering and clustering

Area of Science:

• Digital Image Processing
• Computer Vision
• Image Restoration

Background:

• Total variation (TV) models are effective image priors for regularization-based image processing.
• However, TV models can produce poor results and pseudoedges in flat image regions under high noise intensity due to their preference for piecewise constant solutions.

Purpose of the Study:

• To develop a regional spatially adaptive total variation model to overcome limitations of traditional TV models.
• To enhance image processing by reducing pseudoedges and maintaining image details in noisy flat regions.

Main Methods:

• Extraction of spatial information per pixel.
• Application of two filtering processes to suppress pseudoedges.
• K-means clustering for spatial information weighting and classification.
• Control of regional regularization strength based on cluster centers.

Main Results:

• Effective reduction of pseudoedges in flat regions of images processed with TV regularization.
• Preservation of partial smoothness in high-resolution images.
• Demonstrated robustness against noise intensity variations in super-resolution tasks.

Conclusions:

• The proposed region-based adaptive TV model outperforms traditional pixel-based methods.
• It offers improved robustness by mitigating noise effects on spatial information extraction.
• The model effectively reduces pseudoedges while preserving image details, enhancing overall image quality.