Illumination-Invariant Feature Point Detection Based on Neighborhood Information

Area of Science:

• Computer Vision
• Image Processing

Background:

• Feature point detection is fundamental to computer vision.
• Achieving geometric and illumination invariance in feature detection remains a significant challenge.

Purpose of the Study:

• To propose a novel illumination-invariant feature point detection method.
• To enhance the robustness and stability of feature detection under varying illumination conditions.

Main Methods:

• A two-step approach categorizing feature points based on neighborhood connectivity and pixel distribution.
• Theoretical analysis to demonstrate lower computational complexity compared to existing methods.

Main Results:

• The proposed method outperforms learning-based detection methods in feature point count, matching points, and repeatability rate stability.
• Experimental results show superior illumination robustness compared to state-of-the-art methods, especially under large photometric variations.

Conclusions:

• The developed method offers significant improvements in illumination robustness for feature point detection.
• It provides a more stable and reliable alternative to current feature detection techniques, particularly in challenging visual environments.