+ All Categories
Home > Documents > 1 Vladimir Botchko [email protected] Lecture 7. Image Segmentation Lappeenranta University of...

1 Vladimir Botchko [email protected] Lecture 7. Image Segmentation Lappeenranta University of...

Date post: 20-Dec-2015
Category:
View: 219 times
Download: 2 times
Share this document with a friend
25
1 Vladimir Botchko [email protected] Lecture 7. Image Lecture 7. Image Segmentation Segmentation Lappeenranta University of Technology (Finland)
Transcript
Page 1: 1 Vladimir Botchko botchko@lut.fi Lecture 7. Image Segmentation Lappeenranta University of Technology (Finland)

1

Vladimir Botchko [email protected]

Lecture 7. Image Segmentation Lecture 7. Image Segmentation

Lappeenranta University of Technology (Finland)

Page 2: 1 Vladimir Botchko botchko@lut.fi Lecture 7. Image Segmentation Lappeenranta University of Technology (Finland)

2

Image segmentation

Detection and discountinuities: point, line and edge detection

Edge linking and boundary detection: local and global (Hough Transform) processing.

Thresholding: illumination, global and adaptive thresholding, optimal thresholding.

Region-oriented segmentation Motion-based segmentation

Page 3: 1 Vladimir Botchko botchko@lut.fi Lecture 7. Image Segmentation Lappeenranta University of Technology (Finland)

3

Edge models. Step (right) and ramp (left).

Page 4: 1 Vladimir Botchko botchko@lut.fi Lecture 7. Image Segmentation Lappeenranta University of Technology (Finland)

4

Two ideal edges

Page 5: 1 Vladimir Botchko botchko@lut.fi Lecture 7. Image Segmentation Lappeenranta University of Technology (Finland)

5

Edge nature. The computer graphics image (left), an image with explicit edge structure

(right).

Page 6: 1 Vladimir Botchko botchko@lut.fi Lecture 7. Image Segmentation Lappeenranta University of Technology (Finland)

6

Edge nature. The images where almost each pixel belongs to edges

Page 7: 1 Vladimir Botchko botchko@lut.fi Lecture 7. Image Segmentation Lappeenranta University of Technology (Finland)

7

Edge detection. 3x3 Prewitt operator (close to Sobel).After horizontal mask (upper left), after vertical mask

(upper right), sum for two masks (bottom).

Page 8: 1 Vladimir Botchko botchko@lut.fi Lecture 7. Image Segmentation Lappeenranta University of Technology (Finland)

8

Edge detection. Sobel (left), Prewitt (right)

Page 9: 1 Vladimir Botchko botchko@lut.fi Lecture 7. Image Segmentation Lappeenranta University of Technology (Finland)

9

Edge detection. Marr edge detection (Laplacian of Gaussian).

After convolution with a smoothing filter h(r) of =1.5 (left) and =5 (right). Edge detection is given in next slide.

Page 10: 1 Vladimir Botchko botchko@lut.fi Lecture 7. Image Segmentation Lappeenranta University of Technology (Finland)

10

Edge detection. Marr edge detection.

Zero crossing from above (presented in previous slide) image (right) and the same for left image.

Page 11: 1 Vladimir Botchko botchko@lut.fi Lecture 7. Image Segmentation Lappeenranta University of Technology (Finland)

11

Edge detection. Original image (grains).

Page 12: 1 Vladimir Botchko botchko@lut.fi Lecture 7. Image Segmentation Lappeenranta University of Technology (Finland)

12

Edge detection. Sobel (upper left), Prewitt (upper right), Roberts (lower left), Laplacian

of Gaussian (lower right)

Page 13: 1 Vladimir Botchko botchko@lut.fi Lecture 7. Image Segmentation Lappeenranta University of Technology (Finland)

13

Image segmentation

Detection and discountinuities: point, line and edge detection

Edge linking and boundary detection: local and global (Hough Transform) processing.

Thresholding: illumination, global and adaptive thresholding, optimal thresholding.

Region-oriented segmentation Motion-based segmentation

Page 14: 1 Vladimir Botchko botchko@lut.fi Lecture 7. Image Segmentation Lappeenranta University of Technology (Finland)

14

Hough Transform

Page 15: 1 Vladimir Botchko botchko@lut.fi Lecture 7. Image Segmentation Lappeenranta University of Technology (Finland)

15

Hough Transform

Page 16: 1 Vladimir Botchko botchko@lut.fi Lecture 7. Image Segmentation Lappeenranta University of Technology (Finland)

16

Hough Transform for lines from the previous slide (peaks (highlight dots)

define distance and angle value)

Page 17: 1 Vladimir Botchko botchko@lut.fi Lecture 7. Image Segmentation Lappeenranta University of Technology (Finland)

17

Image segmentation

Detection and discountinuities: point, line and edge detection

Edge linking and boundary detection: local and global (Hough Transform) processing.

Thresholding: illumination, global and adaptive thresholding, optimal thresholding.

Region-oriented segmentation Motion-based segmentation

Page 18: 1 Vladimir Botchko botchko@lut.fi Lecture 7. Image Segmentation Lappeenranta University of Technology (Finland)

18

Thresholding

Page 19: 1 Vladimir Botchko botchko@lut.fi Lecture 7. Image Segmentation Lappeenranta University of Technology (Finland)

19

Thresholding. Illumination role. Left is an original image, right is a nonuniform background estimated from 32x32 pixel

small regions. Background 3-d visualization (lowest)

Page 20: 1 Vladimir Botchko botchko@lut.fi Lecture 7. Image Segmentation Lappeenranta University of Technology (Finland)

20

Thresholding. Illumination role. Left image is background subructed from original. Right image

after contrast enhancement.

Page 21: 1 Vladimir Botchko botchko@lut.fi Lecture 7. Image Segmentation Lappeenranta University of Technology (Finland)

21

Thresholding of color images. Segmented land, sky and sea.

Page 22: 1 Vladimir Botchko botchko@lut.fi Lecture 7. Image Segmentation Lappeenranta University of Technology (Finland)

22

Image segmentation

Detection and discountinuities: point, line and edge detection

Edge linking and boundary detection: e.g. Hough Transform.

Thresholding Region-oriented segmentation Motion-based segmentation

Page 23: 1 Vladimir Botchko botchko@lut.fi Lecture 7. Image Segmentation Lappeenranta University of Technology (Finland)

23

Region-oriented segmentation

Page 24: 1 Vladimir Botchko botchko@lut.fi Lecture 7. Image Segmentation Lappeenranta University of Technology (Finland)

24

Region-oriented segmentation

The image taken from Dr. Seppo Leppäjärvi Ph.D. Thesis

Page 25: 1 Vladimir Botchko botchko@lut.fi Lecture 7. Image Segmentation Lappeenranta University of Technology (Finland)

25

Motion Based Segmentation


Recommended