Edge-Aware Filtering with Local Polynomial Approximation and Rectangle-Based Weighting.

Area of Science:

• Computer Vision
• Image Processing
• Signal Processing

Background:

• Traditional image filtering methods often struggle with preserving fine details and spatial variations.
• Noise in images can significantly degrade the quality and essential information.
• Existing guided filtering techniques may lack efficiency or accuracy in complex scenarios.

Purpose of the Study:

• To develop a novel guided image filtering method.
• To improve the preservation of spatial variations and image details.
• To achieve efficient and robust image filtering performance.

Main Methods:

• Utilizing local polynomial approximation (LPA) with range guidance.
• Implementing a multipoint framework for reliable model regression.
• Developing a spatially flexible weighting scheme for the filtering process.

Main Results:

• The proposed method achieves clearer image boundaries.
• It demonstrates superior performance in recovering spatial variations from noisy images.
• Constant computation complexity ensures efficient implementation.

Conclusions:

• The novel LPA-based guided filtering method offers significant advantages over conventional techniques.
• It excels in applications requiring detail preservation and noise reduction.
• Experimental results validate its outperformance against state-of-the-art methods.