GPU accelerated edge-region based level set evolution constrained by 2D gray-scale histogram.

Area of Science:

• Computer Vision
• Image Processing
• Computational Science

Background:

• The level set method (LSM) is widely used for image segmentation due to its ability to handle complex shapes and topological changes.
• However, the computational expense of traditional LSM limits its application in real-time systems.
• There is a need for efficient and robust image segmentation algorithms.

Purpose of the Study:

• To propose a novel, computationally efficient level set algorithm for image segmentation.
• To reduce the computational complexity of LSM for real-time applications.
• To enhance segmentation performance by integrating multiple image features.

Main Methods:

• A new level set algorithm is developed, integrating edge, region, and 2D histogram information.
• The lattice Boltzmann method (LBM) with a body force is employed to solve the level set equation (LSE), significantly reducing computational complexity.
• The algorithm is implemented on NVIDIA graphics processing units (GPUs) for parallel processing.

Main Results:

• The proposed LSM demonstrates high efficiency and robustness against noise.
• The algorithm is independent of the initial contour placement.
• Experimental results on synthetic and real images validate the method's performance, showing effective segmentation of objects with and without clear edges.

Conclusions:

• The proposed LSM, accelerated by LBM and GPU implementation, offers a fast and parallelizable solution for image segmentation.
• The integration of diverse image information (edge, region, histogram) enhances segmentation accuracy and noise resilience.
• This method provides a significant advancement for real-time image segmentation applications.