5. Image Edges

Edges:

• a image feature
  • shouble be invariant to changes in the image
  • local, meaningful, detectable part
• Definition: significant local changes of intensity in an image
  • typically occur on the boundary between 2 different regions in an image

Invariant features:

• the features that have only little changes after changes of particular conditions
• possible cases:
  • Change scale, move camera closer or farther
  • Change viewpoint, a number of possibilities
    • Rotate the object being viewed in the camera plane
    • Rotate the object being viewed out of the camera plane
    • Translate the object being viewed
  • Change the lighting (darker, lighter, shadows, etc.)
• If we can use this feature to distinguish or classify objects, then it must be discriminatory
  • all speed limit signs have the same shape and color

Edge Detection:

3 steps:

1. Noise smoothing to make the noise does not affect the true edges
2. Edge enhancement to filter edges
3. Edge localization to decide which local maxima in the filters output are edges

use derivatives to do edge detection:

Image Gradient:

• The magnitude of gradient provides information about the strength of the edge
  • edge strength is gradient magnitude
• The direction of gradient is always perpendicular to the direction of the edge
  • edge orientation is gradient orientation

Edge Detection Algorithm:

Noise affects edge detection:

• so before edge detection, we need to smooth first

Sobel Edge Detector:

• an example of smoothing first

non-maxima suppression:

Canny Edge Detection:

similar to simple edge detection but

• localizes by thinning edges to 1 pixel by non-maxima suppression
• uses hysteresis thresholding instead of simple thr esholding

, are gradient components,
is edge strength.
**

Hysteresis Thresholding:

• Define a low and a high thresholds.
• If the gradient at a pixel is above “high” declare it an “edge pixel”
• If the gradient at a pixel is below “low”, declare it a “non-edge-pixel”
• If the gradient at a pixel is between “low” and “high” then declare it an “edge pixel”, iff it is connected to

an “edge pixel” directly or via pixels between “low” and “high”


example:

tradeoff between detection and localization:

• detection: find all important edges
• localization: find images accurately
• we can not simultaneously improve detection and localization in edge detection

Reference:

• wikipedia
• handout of COMP4102: Introduction to Computer Vision from Carleton University School of Computer Science, 2019