Discrete Curvature Representations for Noise Robust Image Corner Detection.

Area of Science:

• Computer Vision
• Image Analysis
• Computational Geometry

Background:

• Corner detection is crucial in image analysis and computer vision.
• Existing curvature calculation methods are sensitive to noise and struggle with closely spaced corners.

Purpose of the Study:

• To develop a discrete curvature representation for improved corner detection.
• To address limitations of existing methods regarding noise sensitivity and corner resolution.

Main Methods:

• Investigated discrete curvature representations of single and double corner models.
• Derived model properties to aid in contour corner detection.
• Developed a new corner detection algorithm based on these properties.

Main Results:

• The proposed method demonstrates high corner resolution, accurately detecting neighboring corners.
• It exhibits reduced sensitivity to local contour variations and noise.
• False corner detection is significantly less likely to occur.

Conclusions:

• The novel discrete curvature approach offers superior performance in corner detection compared to state-of-the-art methods.
• The detector shows robustness against noise and affine transformations.
• It provides enhanced accuracy and reliability for various image analysis tasks.