Learning Methodologies to Generate Kernel-Learning-Based Image Downscaler for Arbitrary Scaling Factors.

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Image downscaling is crucial for adapting content to diverse display resolutions.
• Current methods like bicubic interpolation poorly preserve image details.
• Advanced methods face challenges in hardware cost, target image definition, and training for all factors.

Purpose of the Study:

• To develop a novel kernel-learning-based image downscaler.
• To improve detail preservation quality during image downscaling.
• To support arbitrary downscaling factors using simple linear mapping.

Main Methods:

• A method to generate ideal downscaled targets considering aliasing and detail preservation.
• A training technique using pixel relationships and hierarchical region analysis.
• A kernel-sharing strategy for efficient generation across multiple downscaling factors.

Main Results:

• Demonstrated superior edge preservation, with recall, precision, and F1 scores improved by up to 0.141, 0.079, and 0.053, respectively.
• Achieved excellent edge consistency compared to benchmark methods.
• Received the highest preference in user studies for its simplicity and effectiveness.

Conclusions:

• The proposed kernel-learning downscaler significantly enhances image detail preservation.
• The method offers an efficient and effective solution for arbitrary image downscaling.
• It outperforms traditional and complex methods in both quantitative metrics and user preference.