EE 475

Digital Image Processing

Term Project
Interactive Segmentation Using
Intelligent Scissors
Instuctor: Professor Bulent Sankur
Prepeared by: Yusuf Ziya Isik and Ashat Turlibayev
Intelligent Scissors
Fully automatic segmentation is an
unsolved problem due to wide variety of
images.
Intelligent Scissors is a semi-automatic
general purpose segmentation tool.
The efficient and accurate boundary
extraction, which requires minimal user
input with a mouse, is obtained.

The underlying mechanism for the

Intelligent Scissors is the “live-wire” path
selection tool.
Live wire tool
Boundary detection is formulated as an
optimal path search in a weighted graph.

Optimal boundaries are computed as the

user moves the mouse starting from a
manually specified seed point.
Live-wire segmentation
When the mouse position comes in
proximity to an object edge, a “live-wire”
boundary snaps to, and wraps around the
object of interest.
Input of a new seed point “freezes” the
selected boundary segment, and the process
is repeated until the boundary is complete.
Image as a Weighted Graph
Basic idea is :

Formulate image as a weighted graph

where pixels represent nodes with directed,
weighted edges connecting them to
neighbors.
LIVE WIRE BOUNDARY
SNAPPING
The boundary finding consists of finding the
globally optimal path from the start node to a goal
node.
The optimal path or boundary is defined as the
minimum cumulative cost path from a start node
to a goal node where the cumulative cost of a path
is the sum of the local costs on the path.This is
done by the efficient use of Dijkstra’s graph
searching algorithm.
algorithm
Local Costs
Since an optimal path corresponds to a segment of
an object boundary, pixels that exhibit strong edge
features are made to have low local costs.
costs
The local cost function is a weighted sum of the
component cost functions for each of the
following features:
i) the laplacian zero-crossing, fZ,
ii) the gradient magnitude, fG and
iii) the gradient direction, fD.
Local Costs
Letting l(p,q) represent the local cost for the
directed link from pixel p to a neighboring
pixel q

l(p,q) =wG•f G (q) + wZ•f Z (q) + wD•f D (p,q)

The laplacian zero-crossing feature, fZ(q)
it is a binary edge feature used for boundary
localization.

if IL is the output of the convolution of the image

with a laplacian edge operator
The laplacian zero-crossing feature, fZ(q)

However,a discrete Laplacian image

produces very few actual zero values.

Therefore, if IL(q) has a neighbor with a

different sign, of the two pixels, the one
closest to zero represents the zero-crossing
or the position to localize the boundary.
The gradient magnitude feature, fG(q)

It is a strong measure of edge strength, and is

heavily weighted.

High image gradients will correspond to low costs,

therefore:
The gradient direction.
It adds a smoothness constraint to the
boundary by associating a high cost for
sharp changes in boundary direction.
The gradient direction is the unit vector
defined by the image gradients, Gx and Gy
in the x and y direction, respectively.
The formulation of the gradient direction feature cost
Boundary Detection as Graph
Searching
The graph search algorithm is initialized by
placing a start or seed point, s, with a
cumulative cost of 0, on an otherwise empty
active list, L

All the other points in the image are

initialized with infinite cumulative costs.
Boundary Detection as Graph
Searching
A point p is placed on the active list in
sorted order based on its total or cumulative
cost, g(p).
In each iteration, the point or pixel p with
minimum cumulative cost is removed from
L and “expanded” by computing the total
cost to each of p’s unexpanded neighbors
For each neighbor q of p, the cumulative
cost to q :gtmp = g(p)+l(p,q).

If gtmp< g(q) then g(q) is assigned the new

cumulative cost and an optimal path pointer
is set from q back to p. The process repeats
until al the image pixels have been
expanded.
The cost expansion algorithm
The cost expansion algorithm
Interactive "Live-Wire"
Segmentation Tool
Once the optimal path pointers are generated, a
desired boundary segment can be chosen
dynamically via the free point specified by the
current cursor position.
Interactive movement of the free point by the
mouse cursor causes the boundary to behave like a
live wire as it erases the previous boundary points
and displays the new minimum cost path defined
by following path pointers from the free point
back to the seed point.
By constraining the seed point and free points to lie near a
given edge, the user is able to interactively "snap" and
"wrap" the live-wire boundary around the object of interest.
new seed points
When movement of the free point causes
the boundary to digress from the desired
object edge, input of a new seed point prior
to the point of departure effectively "ties
off" or freezes the boundary computed up to
the new seed point and reinitiates the
boundary detection starting from the new
seed point.
new seed points
Interactive "Live-Wire"
Segmentation Tool
Since each pixel (or free point) defines only
one optimal path to a seed point, a
minimum of two seed points must be
deposited to ensure a closed object
boundary.
Input of a new seed point “freezes” the
selected boundary segment, and the process
is repeated until the boundary is complete.
A minimum of two points are
required
A cursor snap
Placing seed points directly on an object's edge is
often difficult and tedious.
To facilitate seed point placement, a cursor snap is
available which forces the mouse pointer to the
maximum gradient magnitude pixel within a user
specified neighborhood.
The neighborhood can vary from 1x1 (resulting in
no cursor snap) to 15x15 (where the cursor can
snap as much as 7 pixels in both x and y).
Two novel enhancements:
Two novel enhancements to the basic live
wire methodology include boundary
cooling and on-the-fly training.
Data-driven boundary cooling generates
seed points automatically and further
reduces user input.
On-the-fly training adapts the dynamic
boundary to edges of current interest.
Results:
Results:
CONCLUSIONS

Live-wire boundary extraction provides an

accurate and efficient interactive tool for image
segmentation.
The live-wire tool is intuitive to use and can be
applied to black and white or color images of
arbitrary content and complexity.
As a practical matter, live-wire boundary snapping
typically requires less time and effort to accurately
segment an object than it takes to manually input
an initial approximation to the object boundary.