image segmentation - Mahidol UniversityImage segmentation is a task to distinguish the features...

12
Chaiwoot Boonyasiriwat November 9, 2020 Image Segmentation

Transcript of image segmentation - Mahidol UniversityImage segmentation is a task to distinguish the features...

Page 1: image segmentation - Mahidol UniversityImage segmentation is a task to distinguish the features (objects or structures) or the foreground from the background in an image. Here we concentrate

Chaiwoot BoonyasiriwatNovember 9, 2020

Image Segmentation

Page 2: image segmentation - Mahidol UniversityImage segmentation is a task to distinguish the features (objects or structures) or the foreground from the background in an image. Here we concentrate

2

▪ Image segmentation is a task to distinguish the features

(objects or structures) or the foreground from the

background in an image.

▪ Here we concentrate on two approaches: thresholding

and boundary-based segmentation.

Image Segmentation

Russ and Neal (2016, p. 381)

Page 3: image segmentation - Mahidol UniversityImage segmentation is a task to distinguish the features (objects or structures) or the foreground from the background in an image. Here we concentrate

3

▪ Selecting features in a scene can be accomplished by

thresholding brightness values. The resulting image is a

binary or two-level image.

▪ Threshold values may be set based on the histogram.

Brightness Thresholding

Russ and Neal (2016, p. 381-382)

Page 4: image segmentation - Mahidol UniversityImage segmentation is a task to distinguish the features (objects or structures) or the foreground from the background in an image. Here we concentrate

4

Histogram can be interpreted as a sum of 3 Gaussian

peaks representing R, G, and B to colorize an image.

Colorizing an Image

Russ and Neal (2016, p. 384)

MRI image

Page 5: image segmentation - Mahidol UniversityImage segmentation is a task to distinguish the features (objects or structures) or the foreground from the background in an image. Here we concentrate

5

Thresholding of Multiband Images

Russ and Neal (2016, p. 394)

Original image Pixel color values plotted

in biconic HSI space

Resulting binary

image

Selection of HS values

for green candy

Page 6: image segmentation - Mahidol UniversityImage segmentation is a task to distinguish the features (objects or structures) or the foreground from the background in an image. Here we concentrate

6

Image Histogram in Various Spaces

Russ and Neal (2016, p. 395)

RGB Cylindrical HSI Spherical L*a*b*

Page 7: image segmentation - Mahidol UniversityImage segmentation is a task to distinguish the features (objects or structures) or the foreground from the background in an image. Here we concentrate

7

▪ Texture can be used for image segmentation.

▪ Below is an image containing 5 regions with different

texture. The brightness of pixels in each region has a

different probability distribution.

Thresholding from Texture

Russ and Neal (2016, p. 400)

Probability distribution of pixel

brightness in each region

Page 8: image segmentation - Mahidol UniversityImage segmentation is a task to distinguish the features (objects or structures) or the foreground from the background in an image. Here we concentrate

8

(a) Result of applying a variance operator with 4-pixel radius.

(b) Histogram showing that each region has a distinct

brightness.

(c) Each region can be selected by thresholding the histogram

Thresholding from Texture

Russ and Neal (2016, p. 401)

Page 9: image segmentation - Mahidol UniversityImage segmentation is a task to distinguish the features (objects or structures) or the foreground from the background in an image. Here we concentrate

9

Multiple Thresholding Criteria

Russ and Neal (2016, p. 402)

Original image Texture image using variance Intensity image

Page 10: image segmentation - Mahidol UniversityImage segmentation is a task to distinguish the features (objects or structures) or the foreground from the background in an image. Here we concentrate

10

K-Mean Clustering

Russ and Neal (2016, p. 424)

Original 3 arbitrary center

locations assigned

Resulting labels

for the current

centers

The centroids of

the regions are

assigned as the

new centers.

Regions are

redefined and

labeled.

Repeat the

process

Page 11: image segmentation - Mahidol UniversityImage segmentation is a task to distinguish the features (objects or structures) or the foreground from the background in an image. Here we concentrate

11

Region-Growing Methods

https://en.wikipedia.org/wiki/Image_segmentation#Clustering_methods

▪ Assign a set of seeds that marks the objects to be selected.

▪ The pixel with the smallest difference between its intensity

value and a region’s mean is assigned to the region.

▪ The process continues until all pixels are assigned.

Page 12: image segmentation - Mahidol UniversityImage segmentation is a task to distinguish the features (objects or structures) or the foreground from the background in an image. Here we concentrate

▪ J. C. Russ and F. B. Neal, 2016, The Image Processing

Handbook, 7th edition, CRC Press.

References