You are on page 1of 33

Study of Image Segmentation Schemes and Their Application to Microscopic Images

by

S. Jagannath Prusty Rosen Dash

Guided by

Lecturer P. K. Sa

Objective
To study different schemes of

Image Segmentation and their uses in Computer Vision.


microscopic image of a ceramic material and extract required feature.

To segment a

Motivation
In

the course of study of different segmentation schemes, each scheme works well for a specific type of images.

To segment a

microscopic image we cannot directly implement any of them. So we proposed a hybrid segmentation scheme.

Contents:
Line

Detection using Hough Transform. Segmentation using Iterative Thresholding Method. Segmentation using Region Growing Method. Segmentation using MAP. Segmentation of a SEM image.

Hough Tranform method:


Normal representation of

a line xcos + ysin = -900 < < +900


y

- Plane

Implementation
Houghtrans (Input image, dtheta, numpeaks, fillgap, minlength, threshold )

Input image

(a)

(b)

(a) All points represented as sinusoidal curves (b) Selection of peak points

Output Image

Iterative Thresholding Method


Algorithm: f = input image T = final global threshold value
1.
2.

3. 4. 5.
6.

7. 8. 9.

T= 0.5( min(f)+max(f) ) if (f(i,j)>=T) g(i,j) = 1 else g(i,j) = 0 Tn = 0.5 (mean(f(g ==1)) + mean(f(g == 0))) if(abs(T - Tn) > T) T = Tn, goto step 2 end

Implementation

Input Image

Ouput Image

Region Growing Method


Algorithm:
Choose a seed point (p). 2. Consider all the N8(p) points.
1.
1 4 6 2 P 7 3 5 8

3.

Out of eight points selected we find out the point that gives minimum intensity difference between the seed point and it, that should be less than a given value.

Algorithm (Contd.):
4.

5. 6.

The resulted point now becomes the new seed point and repeat the procedure from step 2. This process continues till the condition fails and no more seed point is found. Finally the output image is drawn considering the seed points only.

Implementation

Input Image

Output Image

Segmentation using MAP


In

Bayesian statistics, a maximum a posteriori (MAP) estimate is a mode of the posterior distribution.

MAP is

calculated based on two probabilities value:


Prior probability Likelihood probability

P(Si) = Prior probability i.e., the probability of occurance of that state in the entire image. P(Oj/Si) = likelihood of the outcome Oj given state Si is defined as the transition probabilty from Si to Oj
PSi | O j PO j | Si PSi

PO
n k 1

| S k PS k

Input Image

Output Image

Segmentation of a SEM image


In

Ceramic engineering, Scanning Electron Microscope is used to take images (micrographs) of different materials.
have proposed a segmentation scheme to extract features like counting number of white or black nano particles.

We

SEM Image

Input Image

Histogram after contrast streching

Conclusion
We

have successfully implemented the proposed algorithm with constant guidance of our project guide.

The number of

white nano particles according to the algorithm in the input image is 154, which is quite acceptable.

References

R. C. Gonzalez and R. E. Woods. Digital Image Processing. Addison Wesley, 2nd edition, 1992. Anil K. Jain. Fundamentals of Digital Image Processing. Prentice Hall of India, 2008. B. Chanda and D. Dutta Majumder. Digital Image Processing and Analysis. Prentice-Hall of India, 1st edition, 2002. Hamid Hassanpour and Payam M. Farahabadi. Using hidden markov model for paper currency recognition. Elsevier Expert Systems with Applications, 2009. Pankaj Kumar Sa. On the Development of Impulsive Noise Removal Schemes, 2006

D. Patra P. K. Nanda and A. Pradhan. Unsupervised image segmentation using tabu search and hidden markov random field model. National Conference on Recent Advances in Power, Signal Processing and Control, 2004. Amir Najmi Jia Li and Robert M. Gray. Image classification by a two dimensional hidden markov model. Information Systems Laboratory, EE Dept. Stanford University, 1999 Claudio Becchetti and Lucio Prina Ricotti. Speech Recognition. John Wiley and Sons, 2004.

Thank You

You might also like