Chap. 9: Image Segmentation Jen-Chang Liu, 2005. Motivation Segmentation subdivides an image into its constituent regions or objects Example: 生物細胞在影像序列中的追蹤.

Chap. 9: Image Segmentation

Jen-Chang Liu, 2005

Motivation

Segmentation subdivides an image into its constituent regions or objects

Example: 生物細胞在影像序列中的追蹤

Basics of image segmentation

Input: image => Output: attributes Segmentation of nontrivial image is

one of the most difficult tasks in image processing Ex. Separate the sky, clouds, and

mountains

Two approaches

Based on intensity values discontinuity Partition an image based on abrupt

changes, such as edges

Based on intensity similarity in regions Partition an image into regions that are

similar according to some criteria

Outline Point, line detection Edge detection Thresholding Use of motion in

segmentation

Pixeldiscontinuity

Regionsimilarity

Detection of discontinuity

Three basic types of gray-level discontinuity Points, lines, and edges

Method: run a mask through the image

9

1iiizwR

Input: 3x3 region z1, z2 , … , z9

Output: z5 -> R

z1 z2 z3

z4 z5 z6

z7 z8z9

w1 w2 w3

w4 w5w6

w7 w8w9

Point detection

Point detection mask1. Mask operation over whole image2. Threshold: |R| T

渦輪葉片 X 光 thresholdingpoint mask

Problem with previous point detection

Assume single-pixel discontinuity Bigger region of point?

Assume simple background: homogeneous background Noisy environment ?

Selection of threshold T

Line detection

1. Line detection mask

2. Which orientation?

Thresholding: |Ri| |Rj| T

Ex. A thin line:0 0 01 2 10 0 0

Hori.R1=8

45o

R2=2-45o

R4=2Veri.R3=2

Example: single-line detection

Input: binary image absolute value after –45o mask |R| T

1Detect line that are one pixel thick

2 Isolated “line points”

Outline

Point, line detection Edge detection

Boundary with arbitrary shape Thresholding Use of motion in segmentation

Definition of edges?

Intuition of edge: set of connected pixels that line one the boundary between two regions

Model of digital edge

Two factors affects acquired edges

Blur 模糊化 Sampling rate, illumination condition

zoom

Noise 雜訊

Method of edge detection: derivatives

Edge detectionmethod

Magnitude of 1st

derivatives T

1. Sign determinesDark or light region

2. Zero-crossing

Methods: 1st derivatives 1st derivatives

2-D gradient

y

fx

f

f

Magnitude =

y

f

x

f

y

f

x

f

5.0

22 )()(

Direction = )/

/(tan 1

xf

yf

x

f

y

f

Example: Sobel filterOriginal: 1200x1600

Ver.

Hor.

abs

+abs

Effects of noise

Unwanted Detailed Structure !

1st 2nd

Noisyedges Fail to detect

Reduce noise: SmoothingSmooth with 5x5 average filter

Ver.

Hor.

abs

+abs

Review 2nd derivative

f(x)f(x-1)

1stderivative

2ndderivative

f(x)-f(x-1)

f(x+1)-2f(x)+f(x-1)

Method: 2nd derivative

Edge detectionmethod

Magnitude of 1st

derivatives T

1. Sign determinesDark or light region

2. Zero-crossing

2nd derivative mask

Laplacian

Example: 2nd derivativeoriginal Sobel

GaussianSmoothing

Laplacian(2nd derivative)

LoG Threshold(LoG)Zero-crossing of LoG

Outline

Point, line detection Edge detection Thresholding Use of motion in segmentation

Foundation of thresholding

Idea: object and background pixels have gray levels grouped into two dominant modes

Original image histogram

Foundation of thresholding

Input f(x,y), given threshold T

Tyxf

Tyxfyxg

),( if 0

),( if 1),(

Issues of thresholding

Selection of threshold T ? Complex environment – illumination Multiple thresholds – more than one

object Global threshold Local threshold

Thresholding as a multi-variable function:

g(x,y) = T[ f(x,y), x, y, p(x,y) ]

Adaptive: Depend onposition

Local: local property func.

1. Automatic selection of T

1. Select an initial T Average gray level Mean of max. and min. gray level

2. Segment the image using T T

G1G2

3. Calculate mean of G1 and G2

4. New threshold: T2 = 0.5( + )T2

5. Repeat steps 2~4 until difference in successive T is small

Example: automatically select T

fingerprint

Initial: gray level mean3 iterationsT = 125.4

2. Effects of illumination

Recall: f(x,y)=i(x,y) r(x,y)

Illumination source

scene

reflection

illumination: reflectance:

Example: illumination

Original image

histogram

Illuminationsource

x

histogram

Example: bad histogram

* The gray levels of the object is mixed with background

4. Motivation for adaptive thresholding

histogram

A singleGlobal threshold

Adaptive local thresholding

Subdivide image into blocks

Q: Improperly segmented subimages !

Iterative subdivisionhistogram

subdivision

Outline

Point, line detection Edge detection Thresholding Use of motion in segmentation

Motion as a clue to extract object

Spatial technique

Reference imagef(x,y,1)

next imagef(x,y,2)

time index

Thresholdeddifference image

1 if |d(x,y)| > T0 otherwise

Use more than one images in time: eliminate noise

Reference imageR(x,y)

Image f(x,y,2)

Accumulativedifference image

counter + 1,

Image f(x,y,3)

…

counter

a. if d(x,y) > T positive ADIb. if d(x,y) < -T negative ADIc. if |d(x,y)| > T absolute ADI

d(x,y)=R(x,y)-f(x,y,t)

Example:

Absolute ADI Positive ADI Negative ADI

* Object shape* Location in ref. image