Internal architecture of distributed real time system of image processing and pattern recognition

1
Internal architecture of distributed real time
system of image processing and pattern recognition

• Gostev I. M. Sevastianiv L. A
• MIEM-PFUR Moscow
• 2005

2
Based supposition(1)

• Pattern is some description of object!
• Pattern recognition separation of input object
  in predetermined class under its features or
  characteristics.
• We use contour of object at the heart of
  recognized objects. (It is base of gestalt
  psychology, and base of human perception of
  object).
• Contour have considerable proportion of
  information about graphical object.

3
Based supposition (2)

• Development methodology of graphical pattern
  recognition to invariance to 2D affine transform
  (translation, scaling and rotation ) with
  receiving as result objects coordinates and its
  angle of rotation relatively sample.

4
Plan of based steps from methodology of image
processing and pattern recognition
Preliminary processing
Input criteria and samples
Receive sample contour
Binarisation
Pattern recognition process
Clusterisation
Contour tracing
III
I
II
5

• Delta-segmentation principles
• Using fly window in which calculated statistical
  parameters of signals on based which is
  assignment value of cutting level.
• Using Delta modulation with only two resulting
  value of signal.

Input Image
Different level of intensity
Output image
6
Delta segmentation results (1)
Input image and image after delta - segmentation
PS. Intermediate image processing is absent
7
Compare delta-segmentation to another methods

• Input Image.
• b) Image after processing SUSAN method.
• c) Image after processing Delta segmentation
  method.
• d) Image after processing Canny method.

a)
b)
c)
d)
8
Image contour tracing.
Two fragment of image after step of contour
tracing
Zooming of fragment of image
N.B. Characteristic feature is receive closed
contours of object always.
9
Clustering and samples
Step conclude is Clusters construction
building verbal description of isolated closed
contour of objects and saved its to a file. Any
of objects may use as sample for process of
pattern recognition
Contour of recognition object (Cluster)
Noise on image (Clusters)
Noises object (Cluster).
10
Common method of image processing.
11
Understanding of sample

• Sample this is verbal description of aggregate
  of groups of parameters, which unambiguously
  described a object.
• On such of groups are
• Processing images methods and condition of
  refinement of object
• Coordinate parts of objects
• Pattern recognitions methods
• Classification thresholds in pattern
  recognitions methods

Example of Samples
12
Example of implementation subset of sample
The set
is
primary vectors of properties is as set of
points in 2d space
The set of secondary properties is
Let as named centre of mass element of set
13
Understanding contour function
- sorted function of set
on angle
in order its increasing.
- function of interpolation of points of object
- normalization of contour function.
14
Third step of methodology -
recognition process
Input clusters for pattern recognition
1
System identification is based on consecutive
compare clusters whith preload sample in set of
methods. Each next method is more calculation
difficult and more precision.
2
3
4
Recognized objects
sieve of the methods
15
Method of consecutive weighing.
MCW this is
16
Dimension compare
90
Example of object in rectangular

• Dimension is determinate on the next algorithm
• Find maximum from contour function.
• From this maximum with shift on 900 gets second
  vekue.

17
Common idea of method of geometric correlation

• Next step of algorithm
• Calculate difference value on all 3600 between
  two contour function (current and sample ).

• All difference is summing and dividing on
  numbers of point.
• Made shift on ?t grades and calculate new
  difference.
• If minimal sum is smaller then classification
  tolerance, then object is recognized. The angle
  of shift of this sum is angle of shift object
  relative sample.

18
Geometric correlation 1 (GC1)
as function if difference values
and
Let us
where
Let us function of deviation as
where

The function of recognition on based of
geometric correlation 1 is
where
19
Geometric correlation 2 (GC2)
as mean deviation
Let us function
from
where

The function of recognition on based of
geometric correlation 2 is
where
20
Example of pattern recognition (1)
Sample of objects (zoomed)
Message about result of recognition
Image in real size
Zooming fragment
21
Example of pattern recognition (2)
Sample of objects.
Recognized objects
22
Example of pattern recognition (3)
Sample of objects
23
Parallel processing of input graphical flow
Result of image processing and pattern recognition
Input information flow is cutting on part and
each part is processing into parallel process.
24
Conveyer processing image and recognition in
STIPR2000
Input image
k
k, l, m, n required numbers of line in a method
l
1-st Method
m
2-nd method
n
3-d Method
Output results
N- method
Conveyer of methods image
25
Controller of flow
26
Work diagram of the system for dynamic
identification graphic object (1)
27
Work diagram of the system for dynamic
identification graphic object (2)
28
The work is carried out by the associated
professor Gostev Ivan Michailovich and his
post-graduates students.
??? (095) 916 -8886 mob 7 916-610-7801 Fax
[phone] E-mail igostev_at_gmail.com
igostev_at_list.ru
Address Moscow 117419 Ordzhonikidze 3
Thank You!