Adaptive feature extraction method for capsule endoscopy images

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Traditional oriented FAST and rotated BRIEF (ORB) methods use fixed thresholds, limiting feature discrimination in capsule endoscopy.
• ORB descriptors struggle to effectively distinguish features within capsule endoscopy images, impacting diagnostic accuracy.

Purpose of the Study:

• To introduce a novel feature detection method, adaptive threshold FAST and FREAK in capsule endoscopy images (AFFCEI), for improved capsule endoscopy image analysis.
• To enhance the accuracy and efficiency of feature extraction and matching in capsule endoscopy by addressing limitations of existing methods.

Main Methods:

• Constructs an image pyramid and adaptively calculates pixel thresholds based on local gray value contrast for feature extraction.
• Utilizes the FREAK descriptor to enhance feature discrimination from stomach images.
• Employs grid-based motion statistics and RANSAC algorithm for refined feature matching and outlier rejection.

Main Results:

• The AFFCEI method demonstrated a 5% improvement in average matching score compared to the ASIFT method.
• AFFCEI achieved an average running time that was 4/5 of the ASIFT method, indicating increased efficiency.
• The proposed method effectively tracks features in moving capsule endoscopes, outperforming previous state-of-the-art techniques.

Conclusions:

• AFFCEI offers a more robust and efficient solution for feature extraction and matching in capsule endoscopy.
• The adaptive thresholding and FREAK descriptor significantly improve feature discrimination in challenging endoscopic environments.
• This advancement holds potential for enhanced diagnostic capabilities in capsule endoscopy procedures.