Multi-object geodesic active contours (MOGAC).

Area of Science:

• Computer Vision
• Medical Image Analysis
• Computational Imaging

Background:

• Image segmentation for numerous objects (e.g., cells, brain structures, body parts) is a growing challenge.
• Multi-Object Level Set Methods (MLSM) offer sub-pixel precision but are computationally and memory-intensive.
• Existing methods like region growing and graph cuts are efficient but lack sub-pixel accuracy.

Purpose of the Study:

• To develop a computationally and memory-efficient parallel implementation of Multi-Object Level Set Methods (MLSM).
• To introduce a novel method, Multi-Object Geodesic Active Contours (MOGAC), that overcomes the performance limitations of current MLSM.
• To enable precise segmentation of thousands of objects in 2D and 3D images.

Main Methods:

• A novel parallel implementation of MLSM is presented, leveraging algorithmic sparsity for reduced memory footprint.
• The Multi-Object Geodesic Active Contours (MOGAC) method utilizes a label mask image and unsigned distance field to represent multiple objects.
• The algorithm's time complexity is analyzed as O((M^d)/P) in d dimensions, independent of object count.

Main Results:

• The MOGAC method demonstrates significant improvements in computational and memory efficiency compared to traditional MLSM.
• The algorithm achieves sub-pixel precision in image segmentation tasks.
• Successful application of MOGAC is shown for both 2D and 3D image segmentation problems.

Conclusions:

• MOGAC offers an efficient and precise solution for large-scale multi-object image segmentation.
• The parallel implementation effectively addresses the performance gap in MLSM.
• This method advances the capabilities of automated image analysis in various scientific domains.