Edges Flashcards
What are edges in images
Areas with strong intensity contrasts
General edge detection strategy
Determine image gradient
Mark points where gradient magnitude is particularly large with respect to neighbors
2D difference operators
Prewitt,
Sober,
Roberts
Prewitt difference operator
- 101
- 101
- 101
111
000
-1-1-1
Sobel difference operator
- 101
- 202
- 101
121
000
-1-2-1
Roberts
01
-10
10
0-1
Compass gradient masks
Use eight masks with usual compass directions
Select largest response
Orientation is the direction associated with the largest response
Gradient magnitude- max response
Gradient direction- direction of max response
Properties of derivative masks
Masks have opposite signs
Sum of masks is zero rather than 1 (like smoothing masks)
1st deriv- high absolute values at contrast
2nd deriv- zero-cropping points at contrast
Edge detection steps
1.filtering
(Tradeoff between edge detection and noise reduction)
- Enhancement
Emphasize pixels with sig change in local intensity value - Detection
Thresholding - Localization
Location of edge can be estimated
Why do we use 2 deriv operators?
Laplacian
Thresholding gradient images produces too many edge points
Weakness of laplacian operator
Very sensitive to noise
Can produce double edges
Unable to detect edge direction
(Use zero crossing property for edge location)
Laplacian of Gaussian
Aka Marr-Hildreth
Basically you smooth image with Gaussian than apply a Laplacian
Edges will be at zero crossings
Problems with zero edge finding
Spaghetti effect (closed loops)
More sophisticated to pull off
(Gradient based edge detection is still used more frequently)