Sie sind auf Seite 1von 9

Marr-Hilderth Edge

Detector

It was developed by John F. Canny in 1986


It was a very popular edge operator before canny
published his paper
It is motivated by human vision system

PROCESS:
1. Gaussian:At first the image is to be smoothen
by gaussian filtering which reduces noise in the
image ,suppose, the input image is I and after
gaussian filtering the image formed is S

=mean and =standard deviation,here,=0


Therefore ,smoothened image
S=Gaussian filter f*image I
Gaussian filtering is done by average
weighting
After filtering the smoothen image is gone
through laplacian
Gaussian can be done using one 2-D filter or
2 one-D filter

2. Laplacian: laplacian is second derivative filter


used to find the change in intensity in the
image, as derivative filters are very sensitive to
noise so it first is gone through the gaussian
smoothing .Hence the laplacian of the gaussian
is derived which is also known as LOG operation.

Laplacian can be done using one 2-D filter or


four one-D filter(multiplication complexity
lesser here)

3. Finding zero crossing: we know that if first


derivative is the max second derivative is going
to be zero, based on this concept we find the
zero crossing in order to find the points where
change in intensity is maximum(i.e the points
corresponding to an edge)

Now we have to apply a threshold to mark the


edge .

ADVANTAGES:
Detecting edge and orientation of edge.
Fixed characteristics in all direction
DISADVANTAGE:
Sensitivity to noise
Localization problem

EDGE DETECTION USING LOG

CANNY EDGE DETECTOR


It has more engineering and mathematical
concept. It is an optimal edge detection with
some criteriai. Good detection:It should not
give an edge where there is no
edge and should give an edge
where actually the edge is.
ii. Good localization: the edges
detected should be as close as
possible to the true edges.

iii.

Single response constraint: the


detector must return one point
only for each edge point, should
not have multiple point.
It uses gradient
STEPS:
1.smooth images using gaussian filter: The
gaussian filter can be represented as

And the corresponding equation of gaussian


filter is hereby

2.compute derivative of filtered images:


After smoothing the image the sobel is used in
both
horizontal and vertical
direction to get gradient in horizontal(x
direction) as well as in vertical direction(y
Direction)

3.find magnitude and orientation of the


images:
The magnitude is obtained using the formula

|G| = |Gx| + |Gy|

and the formula for finding edge direction is hereby


theta = invtan (Gy / Gx)
4.applying non-maximum suppression: the main
concept of
this is to suppress the pixels which are not local
maximum.

5.apply hysteresis threshold: there are two


thresholds(high & low),if the pixel gradient is
higher than the upper threshold then ->edge
pixel
If the pixel gradient is low then->non edge pixel
If the pixel gradient is between high and low but
connected to a edge pixel then->edge pixel

ADVANTAGE:
This edge detection is robust to noise
It solves the localization problem also
DISADVANTAGE:
As the computation is little bit complex it
needs moretime

Edge detection by canny

Das könnte Ihnen auch gefallen