Joint infrared target recognition and segmentation using a shape manifold-aware level set.

Area of Science:

• Computer Vision
• Machine Learning
• Image Processing

Background:

• Accurate identification, segmentation, and pose estimation of infrared (IR) targets are critical for various applications.
• Existing methods often struggle with joint optimization, requiring separate or sequential processing steps.

Purpose of the Study:

• To develop novel techniques for the joint recognition, segmentation, and pose estimation of IR targets.
• To formulate the problem within a probabilistic level set framework incorporating a shape-constrained generative model.

Main Methods:

• Utilizing a coupled view and identity manifold (CVIM) model for shape priors.
• Iteratively optimizing a level set energy function under CVIM shape constraints.
• Employing particle swarm optimization (PSO) and gradient-boosted PSO (GB-PSO) for efficient multi-modal optimization.

Main Results:

• The proposed CVIM-based approach enables joint recognition, segmentation, and pose estimation.
• PSO algorithms effectively reduce shape matching costs during inference.
• Gradient-boosted PSO (GB-PSO) demonstrates superior performance compared to recent automated target recognition (ATR) algorithms.

Conclusions:

• The developed probabilistic level set framework with CVIM offers an effective solution for joint IR target analysis.
• GB-PSO provides a computationally efficient and high-performing method for ATR tasks.
• This work advances the state-of-the-art in automated target recognition using infrared imagery.