Dynamic Programming Using Polar Variance for Image Segmentation.

Area of Science:

• Computer Vision
• Image Processing
• Computational Imaging

Background:

• Object size is a key feature in polar dynamic programming (PDP) for image segmentation.
• Unconstrained object size can lead to segmenting extraneous high-gradient regions.
• Existing methods may require training data or struggle with complex shapes.

Purpose of the Study:

• To introduce a novel feature, polar variance, for size-unconstrained image segmentation using PDP.
• To develop a technique for segmenting complex shapes by identifying and expanding low-gradient regions.
• To evaluate the proposed method against established active contour segmentation techniques.

Main Methods:

• Defined polar variance as the variance within a polar region relative to a user-defined origin.
• Integrated a novel approach to segment complex shapes by growing low-gradient areas.
• Conducted comparative experimental analysis against active contour methods.

Main Results:

• The proposed polar variance method enables segmentation of objects with varying sizes without prior training.
• The technique effectively segments complex shapes by leveraging low-gradient region expansion.
• Experimental results demonstrate favorable performance compared to other active contour segmentation techniques.

Conclusions:

• Polar variance offers a robust and versatile feature for PDP-based image segmentation.
• The combined approach enhances segmentation accuracy and robustness across different image types and noise levels.
• This method provides a valuable alternative for segmenting objects of diverse sizes and complex geometries.