An Infrared-Visible Image Registration Method Based on the Constrained Point Feature.

Area of Science:

• Computer Vision
• Image Processing
• Remote Sensing

Background:

• Infrared and visible image registration is challenging due to differing imaging principles.
• Traditional methods struggle with complex feature descriptors, mismatched points, and long computation times.

Purpose of the Study:

• To develop an accurate and efficient method for infrared and visible image registration.
• To overcome limitations of traditional point feature-based registration techniques.

Main Methods:

• Extraction of constrained point features using an object detection algorithm, incorporating semantic information.
• Matching of constrained points using a novel Left Value Rule (LV-rule) to avoid point deletion.
• Calculation of the affine transformation matrix using matched point pairs and an automated accuracy evaluation method.

Main Results:

• The proposed method significantly improves registration accuracy by utilizing semantic information.
• The LV-rule ensures strict matching, eliminating the need to discard mismatched or redundant points.
• The framework demonstrates superior performance in accuracy, speed, and robustness over five state-of-the-art algorithms.

Conclusions:

• The constrained point feature-based method offers an effective solution for infrared and visible image registration.
• The approach provides a more accurate, faster, and robust alternative to existing registration techniques.
• The developed evaluation method aids in automatically assessing registration accuracy.