September 2002 L1.1

1
Feature Detection
Feature Detection
2
Image Features Decisions!
Features such as edges, corners, junction, eyes,
are obtained by making some decision from the
image measurements. Decisions are the result of
some comparison followed by a choice. Examples
(i) if a measurement is above a threshold we
accept, not otherwise (ii) if a measurement is
the largest compared to others, we select it.
3
Decisions Edgels (Edge-pixels and Orientation)
Edge threshold Decision!
Edge orientation Decision!
4
Decisions Edgels (cont.)
Strength of the Edgel
Eliminate some spurious locations. Decision!
The gray level indicates the angle the darkest
one is 0 degrees. The larger is the angle the
lighter is displayed, up to 5p/6.

5
Decisions Local Angle Change
Angle change
where
is the contour curvature multiplied by the arc
length ,
where
A contour segment
y
x
6
Decisions Junctions, Corners
Junction threshold Decision!
Remove (Undo) detection if
Eliminate spurious locations. Decisions!
Examining the values of q where
allow us to characterize the
junctions. For example, when only two value of q
pass the test and
or
suggest a corner. Corners are many times called
L-junctions. If three angles are detected, it may
be a T-junction or an Y-junction. T-junctions
exhibit one region with angle near p, and usually
arise in images due to surface occlusions in a
scene. Four angles suggest a X-junction, and
usually arise in images due to surface
transparency in the scene. Note that this
detector also detects many edgels.
7
Decisions Connecting Edgels, Pseudocode
Algorithm to link edgels. Start with a seed
location (xc,yc)
Contour-Follower(xc , yc) if (Edgel(xc ,
yc ) ? NIL ) Link-neighbors(xc ,
yc ,qmax) Link-neighbors-(xc , yc
,qmax) end
Link-neighbors(xc , yc ,qmax) xn xc ?
xqmax cosqmax yn yc ? yqmax sinqmax
if (Edgel(xn, yn) ? NIL )
Link((xc , yc), (xn, yn))
Link-neighbors( xn , yn ,qmax(xn , yn)
) end
8
Decisions Connecting Edgels, Pseudocode
Link-neighbors(xc , yc ,qmax-c) xn xc ?
xqmax cosqmax yn yc ? yqmax sinqmax
if ((Edgel(xn, yn) ? NIL )(Coherence(xc
, yc xn, yn))) Link((xc , yc), (xn,
yn)) Link-neighbors( xn , yn
,qmax-n(xn , yn) ) end
Coherence(xc , yc xn, yn ) if (

and

) return True else Nil end
9
Threshold Parameters Estimation
We have considered at least four parameters
How to estimate them? One technique is
Histogram partition
Plot the Histogram and find the parameter that
best partition it