wk 3 Flashcards
name the geometric events for edge detection
- surface orientation discontinuity
- depth discontinuity
- colour and texture discontinuity
Name the non-geometric events for edge detection
-illumination change
-shadows
-inner reflections
What are the practical issues common to edge detection?
-There is a trade-off between noise suppression and localisation
-Choice of threshold isn’t very objective, a greater threshold less edges
What are the 3 criteria for an optimal edge detector
- Good Detection: low false positive / negative rates
- Localisation: detected edges must be close to true edges
- Single Response: minimise number of local maxima around true edge
What are the 4 steps (without detail) in canny edge detection
1) First derivative Gaussian applied as X and Y components to the image
2) the magnitude of the X and Y components of the Gaussed image
3) compute the directional matrix of the image
4) Non-maxima suppression
5) hysteresis / double thresholding
How is the magnitude matrix and directional matrix of an image calculated during canny edge detection
1) take partial derivatives w.r.t x and y components of the gaussian filters x and y components
2) convolve x and y first order Gaussian with the image to get G_y and G_x
3) take the magnitude of G_y and G_x by abs sum
4) find directional matrix by arctan(G_y / G_x)
How do you compute non-maxima supression
-convert each element of directional matrix D to one of 4 directions according to its angle. Then for each edge, find other edges in the neighbourhood going in the same direction
- if maximum value: leave as is
- if not maximum value: set to 0
how does hysterisis thresholding work in canny edge detection
after non-maxima surpression:
-set two thresholds t1 and t2 where t2 > t1 and t1 is more or less equal to 2* t2
-if edge is greater than t2 its an edge
-if less than t1 its not an edge
-if in between t1 and t2:
–if linked to edge -> is edge
– if not linked to edge -> not edge
