Corner detection and classification using anisotropic directional derivative representations.

Area of Science:

• Computer Vision
• Image Processing
• Pattern Recognition

Background:

• Corner detection is crucial for image analysis and computer vision tasks.
• Existing methods like He & Yung and CPDA have limitations in detection capability and repeatability.
• Anisotropic directional derivative (ANDD) representations offer a novel way to characterize local image variations.

Purpose of the Study:

• To propose a new corner detector and classifier based on ANDD representations.
• To evaluate the performance of the proposed detector against state-of-the-art methods.
• To assess the classifier's ability to distinguish different types of corners.

Main Methods:

• Utilizing Canny edge detection to obtain edge maps and extract contours.
• Calculating normalized ANDD representations at contour pixels to form a corner measure.
• Applying non-maximum suppression and thresholding for corner identification.
• Developing a corner classifier based on the peak number of ANDD representations.

Main Results:

• The proposed corner detector demonstrates competitive detection capability compared to He & Yung and CPDA detectors.
• The ANDD-based detector exhibits superior repeatability under affine transformations.
• The corner classifier effectively differentiates between simple, Y-type, and higher-order corners.

Conclusions:

• The proposed ANDD-based corner detector offers a robust and repeatable solution for image analysis.
• The ANDD representation provides a powerful tool for both corner detection and classification.
• This approach advances the field of feature detection in computer vision.