A fast and accurate feature-matching algorithm for minimally-invasive endoscopic images

Area of Science:

• Computer Vision
• Robotics
• Medical Imaging

Background:

• Feature matching is crucial for robotic-assisted minimally-invasive surgery (MIS).
• Existing methods often rely on chronological order or organ motion assumptions.
• Unexpected events like occlusions or illumination changes challenge current feature matching.

Purpose of the Study:

• Introduce the hierarchical multi-affine (HMA) algorithm for improved feature matching.
• Enhance accuracy, speed, and robustness in endoscopic image analysis.
• Overcome limitations of existing feature-matching techniques in MIS.

Main Methods:

• Developed the hierarchical multi-affine (HMA) algorithm.
• Utilized appearance-based matching followed by geometric constraint-based outlier removal.
• Tested on a large, annotated dataset of over 100 real MIS image pairs.

Main Results:

• HMA generated a larger number of image correspondences compared to existing methods.
• Demonstrated increased speed, accuracy, and robustness in feature matching.
• Successfully recovered image features after sudden camera events and illumination changes.

Conclusions:

• The HMA algorithm significantly outperforms state-of-the-art methods for feature matching in MIS.
• HMA provides a robust solution for challenging endoscopic scenarios.
• The HMA algorithm and its associated image database are publicly available.