1 Image Segmentation ISANDSP GROUP Pixel Oriented Image

��� 1 ������� Image Segmentation ISAN-DSP GROUP

Pixel Oriented Image Segmentation �������������� ������ Intensity Thresholding T = 102 ��������� ��� Histogram ������ multimo �������������� After thresholding

Image Segmentation based on Histogram Profile �������������� Histogram T 1 = 158 T 2 = 196 T 3 = 228 T 1< P <T 2 T 2< P <T 3 P > T 3

����� Image Segmentation based on Histogram Profile �������������� peak ���� Histogram Image degraded by Gaussian noise (s =12) T 1 = 158 T 2 = 196 T 3 = 228 T 1< P <T 2 T 2< P <T 3 P > T 3

����� Image Segmentation based on Histogram Profile ������������� T=0. 4 Error ������ ISAN-DSP GROUP

Pixel Oriented Image Segmentation for Multispectral Images ����� pixel oriented image segmentation ���� multispectrum ������ partition �� Feature spac ������������ Partition Image Domain Boundary ������ Munich ���� Blue 1 0. 5 0 1 0. 2 0. 8 0. 4 0. 6 From www. jpl. nasa. gov/radar/ sircxsar/munch. html 0. 4 0. 8 Red 1 0. 2 Green Feature Space (RGB)

Region Oriented Image Segmentation �������������� ��������� �������������������� Region oriented image segmentation 1. Region Growing 2. Region Splitting and Merging P Connected pixels Pixel P and its neighbors

Region Splitting and Merging Image Segmentation Region Splitting ISAN-DSP GROUP

Region Splitting and Merging Image Segmentation ���������� Region Spliting ����� standard deviation ��� ����� ISAN-DSP GROUP

Region Splitting and Merging Image Segmentation 2. Merging ����������� ��� ISAN-DSP GROUP

Region Splitting and Merging Image Segmentation ���������� Region Merging ��������� ISAN-DSP GROUP

Gradient Vector Field Gradient ������� Vector ����������� Gradient Vector Field ������� �� ����

Gradient based image segmentation ����� |ÑP| Thresholding ��� T ����� Edge map ��� T �������������� T=60 T=100

Laplacian based image segmentation ����� Ñ 2 P Zero Crossing Detection Edge map Laplacian �������� Zero crossing Detection ����� thresholding ����� T=0 ISAN-DSP GROUP

Gradient VS Laplacian Gradient Laplacian ÑP Ñ 2 P Local Extrema Zero Crossing �������� of |ÑP| ������� (step edge) �������������� ����������� ��� Thresholding Zero Crossing ������� Detection |ÑP| ����������� �� edge map ������ �� ISAN-DSP GROUP

��� 2 Image Segmentation Algorithm Based on Model of A Particle in A Vector Image Field ISAN-DSP GROUP

New Approach : Model of a Particle in a Vector Field Gradient Vector Field Edge Vector Field (Hamiltonian Gradient Field)

Edge Vector Field Edge vector field ��������� ISAN-DSP GROUP

Particle Model in an Edge Vector Field ����������� edge vector field ���� a>0 ������������ : �� edge vector field ����������� (spiral trajecto

Particle Model in an Edge Vector Field ����� P 0 Edge vector field �����

Normal Compressive Vector Field ������������ vector field ������ compressive) ���������� Normal Compressive Field

Normal Compressive Vector Field ISAN-DSP GROUP

Combined Orthogonal Vector Field �������� ISAN-DSP GROUP

Particle Motion in A Combined Orthogonal Vector Field ���������� ���������

Particle Motion in A Combined Orthogonal Vector Field Edge vector field Particle trajectory P 0 Shape distortion at a corner

Boundary Extraction of Multiple Objects 1 -way boundary extraction 2 -way boundary extraction ������������������������� ISAN-DSP GROUP

Boundary Extraction Algorithm Based on Particle Model in a Vector Image Field . 1����� Mask ���������� Gradient ��� Laplacian ������. 2����� edge vector field ��� normal compressive vector f . 3������������ Local maxima ��� Gradient magnitude image. 4������������ . 5 ������������������������� ISAN-DSP GROUP

Experimental Results: MRI Image (Obtained using multiple scale Gaussian differential masks with smin = 0. 5, smax = 2. 0) ������ The whole brain atlas, www. med. harvard. edu/AANLIB/home. html ISAN-DSP GROUP

Experimental Results: Flaw Detection in X-Ray Images Original image Gradient image Boundary extraction results (s=1. 0)

Experimental Results: Multiscale Boundary Extraction Gradient images and extracted boundaries s 1=0. 5 s 2=4. 77 s 3=18. 81

Other Popular Image Segmentation Techniques 1. Level Set Method 2. Active Contour (Snake) 3. Marching Cube 4. Water Shading 5. Nonlinear Diffusion 6. Hough Transform ISAN-DSP GROUP

Image Segmentation : Future Trends. 1 Multiresolution Image Segmentation -Pyramid Representation of Regions -Application of Wavelet Transform to Image segmentation. 2 Intelligent Image Segmentation Algorithm -Incorporating high level knowledge based system for image segmentation -Segmentation and interpretation at the same time ISAN-DSP GROUP

ISAN-DSP GROUP