SurgIPC: a convex image perspective correction method to boost surgical keypoint matching

Area of Science:

• Computer Vision
• Medical Imaging
• Surgical Technology

Background:

• Keypoint detection and matching are crucial for surgical image analysis.
• Existing methods struggle with perspective distortion caused by camera motion.
• Current warping techniques are unsuitable for surgical images due to unrealistic assumptions.

Purpose of the Study:

• To develop a novel method for correcting perspective distortion in surgical images.
• To improve the robustness of keypoint detection and matching under varying camera poses.
• To enable more accurate analysis and reconstruction of surgical scenes.

Main Methods:

• Proposed Surgical Image Perspective Correction (SurgIPC), a linear least-squares (LLS) method.
• Utilized depth maps to warp surgical images, addressing perspective effects.
• Employed conformal flattening theory to preserve angles and minimize resampling artifacts.

Main Results:

• SurgIPC significantly improved the number of correct correspondences in keypoint matching.
• Demonstrated effectiveness on both controlled phantom data and real surgical images.
• Showcased performance gains in downstream tasks like keyframe matching and 3D reconstruction (Structure-from-Motion).

Conclusions:

• SurgIPC effectively enhances keypoint matching in surgical imaging.
• The LLS approach ensures computationally efficient and reliable performance.
• SurgIPC is readily integrable into existing computer-aided surgery systems.