Uncertainty Quantification for Scale-Space Blob Detection

Area of Science:

• Image processing
• Computational mathematics
• Astrophysics

Background:

• Blob detection is crucial for analyzing uncertain images from noisy measurements.
• Existing methods often struggle to accurately represent positional and size uncertainties.

Purpose of the Study:

• To develop a robust method for blob detection in uncertain images.
• To represent blob uncertainty using regions in a 3D scale space.

Main Methods:

• Extending scale-space theory to incorporate uncertainty.
• Utilizing level sets of a total variation functional minimizer within a high-dimensional tube.
• Comparing numerical approaches for solving the non-smooth optimization problem.

Main Results:

• A novel methodology for representing blob uncertainty in scale space.
• Demonstrated effectiveness in deconvolution and astrophysical models.
• Precise and physically interpretable representation of detected blob uncertainty.

Conclusions:

• The proposed method effectively handles uncertainty in blob detection.
• Offers significant improvements for applications in astrophysics and image deconvolution.
• Provides a physically meaningful way to quantify blob uncertainty.