An Optimal Multiedge Detector
for SAR Image Segmentation
Yao WU
12/02/2009
Outline
• Introduction
• Research Design and Methods
• Implementation
Introduction
• Feathers of synthetic aperture radar (SAR)
image
(1) speckle = multiplicative noise
(2) low SNR
• Requirements for edge detector:
(1)localization precision
(2)speckle suppression
Research Design and Methods
• A. Multiedge Model
• B. Linear MMSE (minimum mean square
error) Filter
• C. ROEWA Operator
Generate a line of SAR image
1) line of non noisy image R(x)
Poisson distribution (lamda = 50)
Generate a line of SAR image
2) multiplicative speckle noise n(x)
gamma distribution
Generate a line of SAR image
2) multiplicative speckle noise n(x)
gamma distribution
Generate a line of SAR image
3) Line noisy image I(x)
Spectral analysis of a line of SAR
image
1) periodogram
Spectral analysis of a line of SAR
image
2) Correlogram
Detection of breaks in a line of SAR
image
1) Generate ISEF (infinite symmetric
exponential filter) filter
Filter size = 20 Filter size = 3
Image value {1,2} Image value {0, …,
255}
Detection of breaks in a line of SAR
image
2) Denoising and rupture detection for a line
of image with value{1,2}
Detection of breaks in a line of SAR
image
3) Denoising and rupture detection for a line
of image with value{0, …, 255}
Detection of breaks in a line of SAR
image
μ X 1 ( x, y ) = f1 ( x) ∗ ( f ( y ) o I ( x, y ))
ˆ
μ X 2 ( x, y ) = f 2 ( x ) ∗ ( f ( y ) o I ( x, y ))
ˆ
Here ∗ denotes convolution in the horizontal direction and
o
denotes convolution in the vertical direction.
⎧μ μ ⎫
ˆ ˆ
rmax = max ⎨ 1 , 2 ⎬
ˆ 2 μ1 ⎭
⎩μ ˆ
r2− D max ( x, y ) = rX max ( x, y ) + rY2max ( x, y )
2
Detection of breaks in test
images
Detection of breaks in test
images