GYTE

1
GYTE COMPUTER DEPARTMENT PH.D SEMINAR
II PARTICULAR ISSUES IN IMAGE PROCESSING
By Ali Iskurt 2004 Haziran
2
PARTICULAR ISSUES IN IMAGE PROCESSING

• CONTENTS
• SNAKES ACTIVE CONTOURS
• RELATED ARTICLES BOOKS
• THEORY
• IMPLEMENTATION
• DISCUSSION
• CASE STUDY INTRAVASCULAR ULTRASOUND IMAGING
• RELATED ARTICLES
• THEORY
• IMPLEMENTATION
• TESTS
• FURTHER RESEARCH

3
Segmentation by Active Contours RELATED ARTICLES

• Snakes Active Contour Models by Kaas, Witkin
  and Terzopoulos
• A Fast Algorithm for Active Contours and
  Curvature Estimation by Williams and Shah.
• Feature extraction and Image Processing Book
  written by Mark Nixon and Alberto Aguado,

4

• Energy function

Alpha Econtinuity Beta Ecurvature Gamma
Egradient
5

• Econtinuity refers to the distance between a
  pixel and its neighboring Snake Points with
  respect to all the Snake Points. It is calculated
  by subtracting the average distance(between all
  Snake Points) from the distance between the pixel
  and the previous Snake Point.
• Ecurvature refers to the amount curvature between
  a pixel and the neighboring Snake Points. This
  can be calculated using the second derivative,
  which corresponds to
• (next Snake Point - pixel) - (pixel - previous
  Snake Point)
• Egradient merely refers to negative the size of
  the gradient of the pixel.

6
Pseudo-Code of Greedy algorithm of ShahWilliams
7
(No Transcript)
8
(No Transcript)
9
CASE STUDY INTRAVASCULAR ULTRASOUND IMAGING
10
Original image Sobel edges
11
Wall and Plaque Border Detection Algorithm

• Removal of Calibration Markers
• Definition of Region of Interest
• Edge Detection and Coordinate Transformation
• Lamina Border Detection graph
• Plaque Border Detection graph
• Edge Direction

12
Identification of Vessel Wall and Plaque Borders

• Border detection as graph searching
• Identification of External elastic lamina
• Identification of Internal elastic lamina
• Identification of plaque border
• Quantitative evaluation of arterial morphology

13
Basic borders to be detected

• El leading edge of the IVUS response
  represents location of external lamina
• Et trailing edge of echo response to
  external lamina.
• Il leading and trailing edges of the echo
  response depicting the internal lamina.

14
Basic variables

• f(k , i) edge strength of ith node on the
  kth graph profile of the resampled edge subimage
  for laminae detection.
• d (k , i) edge direction
• cE (k , i) cost assigned to a node on path
• pE (k , i ) edge triplet or pattern cost and
  direction by which the node is reached.

15
Cost calculation

• pE (k , i ) max P (x,y) - P (k,i) x ?
  X , y ? Y
• P (k,i) El(k,i) Et(k,i) It(k,i)
• El(k,i) f (k,i) for d(k,i) ?
  -? / 2 , ? / 2
• f( k,i) dp otherwise
• Et(k,i) max f Et (k, ij) where j 3, 6
• It(k,i) max f It (k, ij) where j -4,
  -2
• f Et (k, ij ) f (k,i) for
  d(k,i) ? ? / 2 , 3? / 2
• f( k,i) dp
  otherwise
• f It (k, ij) f (k,i) for
  d(k,i) ? ? / 2 , 3? / 2
• f( k,i) dp
  otherwise

16
Identification of Internal elastic lamina

• cI (k , i) pI (k , i)
  if 4 ( i -j) 11
• pI (k , i)
  sp otherwise
• pE (k , i ) max f Il(x,y) - f Il
  (k,i) x ? X , y ? Y with
• f Il (k,i) f (k,i)
  for d(k,i) ? -? / 2 , ? / 2
• f( k,i) dp
  otherwise

17
IMPLEMENTATION

• rmin 70 ( The beginning radius of ROI.)
• rmax 125 ( The outer boundary radius of
  ROI.)
• r rmax-rmin
• cx215 ( x center of picture)
• cy 200 ( y center of picture)

18
ROI
19
Resampling
20
(No Transcript)
21
Directions
22
External Lamina Detection
23
Display of A Border Corner
24
All three borders detected
25
Correction of Contour points in detected border

• 1. using wide points visualization
• 2. Averaging the distributed points for
  smoothing
• 3. using particular 10degree interval points to
  visualize
• 4. However the above techniques do no
  improvement to gaps or chasms ,thus find average
  distance from origin and points far away will be
  equaled to that average. This will bring some
  resistable and unavoidable error by the way.
• 5. After defining an approximate Contour , use
  snake technique and give high weighting to
  curvature so as to bring points closer to each
  other.

26
technique 3
technique 4
technique 4
technique 4
technique 4
27
Another Method for detecting Plaque Border
USING SNAKES NONLINEAR FILTERING
28
Defining a new Tensor in Coherence-Enhancing
Normal
Perpendicular direction Along edge
?Y
?X
29
After nonlinear filtering
30
CURVATURE technique
Suppress catheter
find_islands along edges
Now we can use Snakes on cleaned image