Feature Preserving Image Smoothing Using a Continuous Mixture of Tensors

Area of Science:

• Computer Vision
• Image Processing
• Scientific Imaging

Background:

• Preserving local discontinuities, terminations, and bifurcations is crucial for many image analysis tasks.
• Feature-preserving denoising is a significant challenge in image processing.

Purpose of the Study:

• To present a novel technique for denoising images while preserving complex oriented structures.
• To develop a method capable of maintaining junctions, corners, and bifurcations.

Main Methods:

• A two-stage process involving pre-processing with a steerable Gabor filter bank to extract local orientation information.
• Representing orientation distribution using a continuous mixture of Gaussians, linked to the Laplace transform of a mixing density (mixture of Wisharts).
• Feature-preserving denoising achieved through iterative convolution with the derived Rigaut type function.

Main Results:

• Experimental results demonstrate superior performance on noisy data, real 2D images, and 3D MRI plant root data.
• The proposed technique effectively preserves complex structures like bifurcating roots.
• Comparison with state-of-the-art anisotropic diffusion filters shows improved feature preservation.

Conclusions:

• The novel technique offers effective feature-preserving denoising for complex oriented structures.
• The method shows promise for applications in computer vision, image processing, and biological imaging.
• The approach outperforms existing methods in preserving image details and structures.