Edge Tracing Using Gaussian Process Regression.

Area of Science:

• Computer Vision
• Machine Learning
• Image Processing

Background:

• Edge tracing is crucial for image segmentation and analysis.
• Existing methods often struggle with image artifacts and occlusions.
• A robust and adaptable edge tracing algorithm is needed.

Purpose of the Study:

• To introduce a novel edge tracing algorithm using Gaussian process regression.
• To develop an algorithm robust to image artifacts and occlusions.
• To demonstrate the algorithm's versatility across different imaging domains.

Main Methods:

• An edge-based segmentation algorithm models edges using Gaussian process regression.
• A recursive Bayesian scheme iteratively searches for edge pixels.
• Combines local image gradient information with global structural information from posterior curves.
• Hyperparameters are tunable and optimizable without prior training.

Main Results:

• The algorithm successfully traces edges by accumulating pixels, converging to the edge of interest.
• Demonstrated robustness to artifacts and occlusions due to uncertainty quantification.
• Efficiently traces edges in image sequences using prior information.
• Validated through applications in medical and satellite imaging, outperforming existing methods.

Conclusions:

• The proposed Gaussian process regression-based edge tracing algorithm offers a robust and adaptable solution for image segmentation.
• Its uncertainty quantification and tunable hyperparameters make it suitable for diverse imaging applications.
• The method shows promise for efficient edge tracing in image sequences.