Detection of intensity changes with subpixel accuracy using laplacian-gaussian masks.

Area of Science:

• Computer Vision
• Image Processing
• Computational Photography

Background:

• Accurate edge detection is crucial for image analysis.
• Existing methods often use approximations like Difference of Gaussians (DOG).
• Subpixel accuracy and efficient processing are key challenges.

Purpose of the Study:

• To develop a novel system for precise edge and line detection in grayscale images.
• To improve upon existing edge detection techniques by enhancing accuracy and efficiency.
• To explore the relationship between image resolution and processing time for edge detection.

Main Methods:

• Utilizing Laplacian-of-Gaussian (LoG) masks for edge detection via zero-crossings.
• Decomposing LoG masks into a sum of two separable filters for exact implementation.
• Employing the facet model for achieving subpixel accuracy in edge localization.
• Investigating edge detection at different image resolutions to optimize processing time.

Main Results:

• The system accurately detects edges with subpixel precision.
• The exact decomposition of LoG masks offers an advantage over DOG approximations.
• Processing edge detection at reduced resolutions significantly decreases computation time with minimal accuracy loss.
• Edges are successfully linked into lines using a described technique.

Conclusions:

• The proposed system provides a robust and efficient method for subpixel edge and line detection.
• Exact LoG mask decomposition and resolution scaling offer practical benefits for image processing.
• This approach advances the field of computer vision by improving accuracy and speed in fundamental image analysis tasks.