Improved Recursive Geodesic Distance Computation for Edge Preserving Filter.

Area of Science:

• Computer Vision
• Image Processing
• Signal Processing

Background:

• Recursive filters based on geodesic distance affinity typically use two 1D recursions, leading to theoretical drawbacks and artifacts in 2D extensions.
• Existing methods struggle with accurate 2D approximations of geodesic distance filters.

Purpose of the Study:

• To propose a novel maximum influence propagation method for approximating the 2D extension of geodesic distance-based recursive filters.
• To improve the accuracy and performance of geodesic distance filters for image denoising applications.

Main Methods:

• Developed a maximum influence propagation method to approximate the 2D extension of geodesic distance filters.
• Implemented an improved recursive filter for image denoising.
• Evaluated the filter's performance against state-of-the-art denoising methods.

Main Results:

• The proposed method partially overcomes the drawbacks associated with the 1D recursion approach.
• The improved recursion demonstrates a better approximation of the true geodesic distance filter.
• Experimental results show comparable or better denoising performance than existing methods, with significantly faster computation (O(8P)).

Conclusions:

• The maximum influence propagation method offers a viable solution for approximating 2D geodesic distance filters.
• The enhanced filter provides superior image denoising results compared to previous recursive implementations.
• The algorithm achieves high efficiency, making it suitable for practical applications.