DIGITAL IMAGE PROCESSING (COM-3371)

Week 7 – February 25, 2002

Topics:

• Invited lecture: Dr. Simon Warfield (http://www.spl.harvard.edu/~warfield ) will talk on

"Exploiting Atlases in Medical Image Segmentation."

• Image representation and description

• Representation schemes
• Chain codes
• Polygonal approximations
• Signatures
• Boundary segments
• The skeleton of a region
• Line segmented encoding

• Boundary descriptors
• Basic descriptors
• Fourier descriptors

• Regional descriptors
• Basic descriptors
• Topological descriptors
• Texture
• Morphology

• Term project - questions, problems, discussion

Readings:
• Chapter 11 (11.1 –11.3) and Chapter 9 (9.1-9.3. 9.5.1, 9.5.2) of text

1
 IMAGE REPRESENTATION
AND DESCRIPTION
INTRODUCTION

Region representation:
- based on external characteristics (its boundary)
- based on internal characteristics (pixels comprising the region)
Region description:
- boundary descriptors, such as boundary length, diameter, curvature, etc.
- regional descriptors, such as area, perimeter, compactness, mean value, etc.

Generally, an external representation is chosen when a primary focus is on

shape characteristics. An internal representation is selected when a primary focus
is on reflectivity properties, such as color or texture.

REPRESENTATION SCHEMES

The segmentation techniques provide results in form of pixels along the

boundary or pixels contained in a region:

It is a standard practice to use boundary representation schemes that

compact the data into representations that are more useful in the computation of
descriptors. One of the schemes are chain codes.

2
Chain codes
• represent a boundary by a connected sequence of straight-line segments of
specified length and direction
• the direction of each segment is coded by using a numbering scheme:

• a typical chain code is generated by following a boundary in a clockwise

direction and assigning a direction to the segments connecting every pair of
pixels
- this method results in quite long chain codes
- any small disturbances along the boundary may cause changes in
code that may not be related to the shape of the boundary

• another technique is to select a larger grid spacing, assign boundary points to

each node of the large grid (depending on the proximity of the original
boundary to that node), and represent the resampled boundary by a 4- 8-code:

3
the boundary representation in figure (c) is: 0033333323221211101101
the boundary representation in figure (d) is: 076666553321212

The accuracy of the resulting code depends on the spacing of the sampling grid.

4
Polygonal approximations

• a digital boundary can be approximated by a polygon

• minimum perimeter polygons
- enclose a boundary by a set of concatenated cells and produce a minimum
perimeter that fits the cell strip:

- merging techniques (merge points along a boundary until the least square
error line fit of the points merged exceeds a preset threshold; repeat the
procedure for the new points along the boundary; at the end of the procedure
the intersections of adjacent line segments form the vertices of the polygon)

- splitting techniques (subdivide a segment successively into two parts until a

given criterion is satisfied -- for example, a requirement might be that the
maximum perpendicular distance from a boundary segment to the line
joining two end points does not exceed a preset threshold)

5
Signatures
• a simple functional representation that can be used to describe and reconstruct
the boundary with appropriate accuracy
• the simplest signature is a plot of the distance from the centroid to the
boundary as a function of angle:

6
 • signature is an translation-invariant representation which allows an object to be
compared with a standard prototype by cyclically shifting the signature of one
with respect to the other in steps, while checking for the best match

Boundary segments

• convex hull technique

- convex hull H of the arbitrary set S is the smallest convex set containing
S; the set H-S is called convex deficiency D of the set S

- the region boundary can be partitioned by following the contour of S and

marking the points at which a transition is made into or out of a component
of the convex deficiency

- in practice, the boundary is usually smoothed prior to partitioning (for

example, by using a polygonal approximation)

7
The skeleton of a region

• medial axis transformation (MAT) generates a "skeleton" of a region

• MAT algorithm:
(1) for each point in the region we find its closest point in boundary,
(2) if a point has more than one such a neighbor --> a point belongs to
the medial axis (skeleton) of the region

• the results of MAT operation depend on the distance measure:

pixel coordinates
p (x,y)
q (s,t)

Euclidean distance between p and q is defined as:

De(p,q)=[(x-s)2 + (y-t)2]1/2

D4 distance:
D4(p,q)=|x-s|+ |y-t|
For example, the pixels with D4 distance ≤ 2 from pixel (x,y) form the
following contours of constant distance:
2
2 1 2
2 1 0 1 2
2 1 2
2

8
D8 distance:
D8(p,q)=max (|x-s|, |y-t|)

For example, the pixels with D8 distance ≤ 2 from pixel (x,y) form the
following contours of constant distance:
2 2 2 2 2
2 1 1 1 2
2 1 0 1 2
2 1 1 1 2
2 2 2 2 2

• MAT operation is computationally expensive

9
BOUNDARY DESCRIPTORS
Length - number of pixels along the contour

Xmax, Xmin, Ymax, Ymin - maximum and minimum coordinates of the

contour

Xcentroid, Ycentroid - X and Y coordinates of the center of the region

Boundary diameter - maximum distance between two points on the

boundary

Curvature - rate of change of slope (example: the difference

between the slopes of adjacent boundary segments)

Shape order (chain count) - number of elements in the chain

Shape number - first difference of smallest magnitude

p. 496

10
11
Fourier descriptors

• the Fourier transform of a boundary representation (chain code,

signature, complex boundary function) is an alternative representation of
the region's shape

• in many cases one can lowpass filter the boundary function spectrum
without destroying the characteristic shape of the region --> this means that
only the amplitudes and phases of the low frequency components in the
spectrum (i.e. the low-order Fourier coefficients) are required to
characterize the basic shape of the object and they can be used as shape
descriptors

12
REGIONAL DESCRIPTORS

Perimeter - length of region's boundary

Area - number of pixels contained within its boundary

Shape factor - Perimeter2/Area

Range - the difference between the largest and the smallest pixel
values in a region

Median - median of pixels' intensity values within the region

13
Mean - average of pixels' intensity values within the region
N
∑ xi
µ = i=1 =x
N

Variance - the second moment about the mean

N
∑ (xi − x)2
σ 2 = i=1
N −1

Standard deviation - square root of a variance

SD = σ 2

Coef. of variation - a ratio of standard deviation to the mean

σ
C=
µ

Skewness - the third moment about the mean; an indicator of

asymmetry in a pixel distribution
N
∑ (xi − x)3
sk = i=1
(N − 1)σ3

Kurtosis - the fourth moment about the mean; relates to the

degree of peakedness or flatness of a distribution
(for the normal distribution kurtosis =3)
N
∑ (xi − x)4
kurt = i=1
(N − 1)σ 4

14
Topological descriptors

• topology - the study of properties of a figure that are unaffected by any

deformation except tearing or folding

• topological features give a global description of a region

examples:
number of holes
number of connected components
number of object edges
number of faces
number of vertices
Euler number (the difference between the number of connected
components and number of holes)

15
p. 505, fig. 8.19

Texture

• texture, as observed in wood grain, stone, cloth, grass, etc., is an important

region description
• no formal definition; intuitively this descriptor provides measures of
properties, such as smoothness, coarseness, regularity, etc.

16
• three principal approaches used to describe the texture are:
statistical - smoothness, coarseness, graininess (using moments,
entropy, etc.)
structural - arrangement of image primitives, such as regularity of
parallel lines
spectral - based on properties of Fourier spectrum (periodic patterns
in an image)

17
 MORPHOLOGY
• Morphological operators - tools for extracting image components that are
useful in the representation and description of region shape (examples:
erosion, dilation, etc.)

• The language of mathematical morphology is set theory

• Erosion - removes pixels from the periphery of a region (it also removes
single pixels)

• Dilation - adds a layer of pixels around a periphery of a region (it also fill
small holes within regions)

18
 If Abi are translations of the binary image A by 1 pixels of the binary image
B, then the union of the translations of A by the 1 pixels of B is called the dilation
of A by B.

The erosion of a binary image A by a binary image B is 1 at a pixel p if and

only if every 1 pixel in the translation of B to p is also in A

19
20
• Erosion and dilation are often used in filtering the images - if the nature of
noise is known, then a suitable structuring element can be used and a
sequence of erosion and dilation operations can be applied for removing the
noise

• Opening - a combination of an erosion followed by a dilation (opening up the

spaces between touching regions, removing pixels in regions which are too
small to contain the structuring element)

• Closing - a combination of an dilation followed by an erosion (fusing narrow

brakes, eliminating small holes, filling gaps smaller than the structuring
element)

21
• The structuring element (probe) does not have to be compact or regular - it
can be any pattern of pixels
• Morphological operations can be used for boundary extraction:

Set A

A eroded by B Boundary extracted by taking the set

difference between A and its erosion

• Morphological operations can also be used for region filling and for
extraction of connected components

• Morphological operations can be used for optical character recognition:

a. Create a model for each character:
- Extract the character to be recognized
- Use expanding or closing to fill holes and cavities
- Shrink the character image to remove unwanted regions and to
reduce the size so that it will fit inside an instance of the character
b. Preprocess the character image (fill the holes, remove unwanted
pixels)
c. Use the character model as a structuring element and perform erosion
d. Compute the connected components
e. Apply the size filter to discard regions that are too small
f. Compute the position of each region that passes through the size filter
--> this provides the position of each recognized instance of the
character model in the image

22