On the extraction and alignment of image edges

Area of Science:

• Computer Vision
• Human Visual Perception

Background:

• Understanding how humans perceive and align image features is crucial for developing advanced image processing algorithms.
• Previous research has explored various image processing techniques, but their alignment with human visual capabilities requires further investigation.

Purpose of the Study:

• To identify image processing techniques (filtering, segmentation, correlation) that mimic human edge alignment in 2D images.
• To evaluate the effectiveness of these processes in matching extracted image edges to original image content.

Main Methods:

• Computed edge-only image representations using pre-filtering and segmentation.
• Applied Laplacian operators to extract luminance boundaries.
• Utilized derivative matched filtering (cross-correlation) to match edge-only images with originals.

Main Results:

• Derivative matched filtering accurately predicts the matching of edge-only images to original images.
• A bias was observed towards matching edges located in the middle-frequency range of the images.

Conclusions:

• Image alignment by human observers can be modeled using derivative matched filtering.
• The human visual system may prioritize middle-frequency edges during image alignment tasks.