Effective Five Directional Partial Derivatives-Based Image Smoothing and a Parallel Structure Design.

Area of Science:

• Computer Vision
• Image Processing
• Scientific Computing

Background:

• Image smoothing is crucial for tasks like segmentation and object classification.
• Existing methods struggle with removing small details and smoothing discrete regions effectively.
• There's a need for faster and more accurate image smoothing techniques.

Purpose of the Study:

• To propose an effective and efficient image smoothing scheme.
• To reduce computational complexity through parallel processing.
• To evaluate the proposed scheme's performance objectively.

Main Methods:

• A novel smoothing scheme combining gradient, Laplacian, and diagonal derivatives.
• Implementation of a parallel processing structure on a graphics processing unit (GPU).
• Objective evaluation using linearly quantized images to assess reconstruction accuracy.

Main Results:

• The proposed scheme demonstrated superior smoothing performance and higher objective scores compared to existing methods.
• It successfully preserved critical edges while removing trivial details.
• GPU implementation resulted in significantly lower computational complexity (18x vs. CPU, 16x vs. L0-based).
• Noise reduction tests showed performance exceeding state-of-the-art by over 5.4 dB.

Conclusions:

• The proposed weighted derivative-based image smoothing scheme offers significant improvements in speed and accuracy.
• Its parallel GPU implementation drastically reduces computational complexity.
• This method is a valuable tool for efficient image smoothing in various applications.