Introduction to The 3rd MMS

1
SRSIE
Semi-automatic Extraction Technique of
Residential Area in High Resolution Remote
Sensing Image
SU Junying, Doctor Candidate Prof. Zhang
Jianqing HU Qingwu , Doctor Candidate
School of Remote Sensing and Information
Engineering Wuhan University, P.R. China
Oct. 20, 2003
MIPPR 2003, Beijing
2
SRSIE
Contents

• Introduction and Background

• Methodology

• Experiment and Results

• Conclusion and Discussion

MIPPR 2003, Beijing
3
SRSIE

• Introduction and Background

• What s Residential area?

Definition Residential area is a spatial entity
consist of buildings, road inside the buildings,
green ground, activity place and factory and it
has some structure, function and spatial
distribution.
Types urban residential area and country
residential area
Urban Residential Area
Country Residential Area
MIPPR 2003, Beijing
4
SRSIE
Background

• Why we need Residential Region information?

• Residential area planning and decision-making in
  economic region planning

• Provide disaster estimation department with
  space distribute information

• Upholding the supervision for living area
  condition

• Aim at Large scope semi-automatic or automatic
  mapping with satellite image especial high
  resolution remote sensing image

MIPPR 2003, Beijing
5
SRSIE
Background

• Is it possible to extract residential area in the
  Remote Sensing image?

• High-resolution image can show the object
  information such as structure, texture and detail
  clearly

• More and more high resolution remote sensing
  satellite

MIPPR 2003, Beijing
6
SRSIE
Background

• Current progress of Residential area extraction

• Mainly Focus on the line object extraction in
  the high resolution images

• Using spectrum characters and classification
  methodology

• Low precision level (69.1? and 52.3? )

• Proposed Methodology

• Using texture character

• Semi-automatic extraction

MIPPR 2003, Beijing
7
SRSIE

• Methodology

• Residential area in High resolution remote
  sensing image

• Mixed with dark and light pixels

• Complex structure

• Difficult to identify background
• and foreground

• Road and Road in the residential area

High Resolution Remote Sensing Image with
Residential Area
MIPPR 2003, Beijing
8
SRSIE
Residential Area
Total Image
Summary
(1) Most of residential area pixels are mixed
pixels and less are pure pixels
(2) The automatic extraction of the residential
area has many difficulties related to the current
research progress in this field
(3) Its difficult to segment with directly grey
information
MIPPR 2003, Beijing
9
SRSIE
Methodology

• Texture Analysis of Residential High resolution
  image

(1) Texture Analysis Method

• Deviation Texture Analysis

• Fourier Texture Analysis

• Grey Symbiotic Texture Analysis

MIPPR 2003, Beijing
10
SRSIE
Methodology
(2) Texture Character Comparisons
View Analysis

• The deviation feature image has high contrast
  between the residential area and the other image
  object

• Fourier texture image is simulate to the original
  image with low contrast

1a Original Image
1b Deviation Texture Image

• The grey symbiotic matrix texture feature image
  has little effect to the original image

1c Fourier Texture Image
1d Symbiotic Matrix Texture Image
MIPPR 2003, Beijing
11
SRSIE
Methodology
(2) Texture Character Comparisons
Statistic Analysis
For the deviation texture image, its easy to
obtain the segmentation threshold for the
residential area extraction just using the grey
histogram statistic data
MIPPR 2003, Beijing
12
SRSIE
Methodology

• Self-adaptive segmentation threshold Using Gauss
  Blur

Gauss Blur
s3
s5
Function Blur the detail and Make same type
object image ( background and residential area)
has the similar grey statistic feature which is
useful to the segment threshold value selection
MIPPR 2003, Beijing
13
SRSIE
Methodology
View Analysis of Gauss Blur
2c Gauss Blur Deviation Image
2b Gauss Blur Image
2a Original Image
(1) Both the detail of background and residential
area are blurred
(2) gauss blur make the deviation texture image
consistency grey distribution inside the
residential area and high contrast to the
background. Thus the self-adaptive threshold can
be easily obtained and the extraction can be well
done.
MIPPR 2003, Beijing
14
SRSIE
Methodology
Statistic Analysis of Gauss Blur
(1) Gauss blur can further increase the
difference values of statistic data between the
residential area image (foreground ) and the
total area image (background).
(2) Gauss blur makes the adaptive segment
threshold possible
MIPPR 2003, Beijing
15
SRSIE
Methodology
Statistic Analysis of Gauss Blur
MIPPR 2003, Beijing
16
SRSIE
Methodology

• Processing flow of residential area extraction

MIPPR 2003, Beijing
17
SRSIE

• Experiment and Results

• Experiment Data

Nine high resolution images with typical
residential area (country residential area, urban
residential area, rarefied and concentrated
residential area) are selected for extraction
experiment and the resolution is ranged from 3m
to 15m.
MIPPR 2003, Beijing
18
SRSIE
Experiment and Results

• Evaluation of Residential Area Extraction

• Method and Steps

(1) Calculate semi-automatic extraction area
M1pixels and actual area M2 pixels manually
selected of the residential area (2) Superpose
M1and M2 and measure the right extraction pixels
number, the error extraction pixels number and
the miss extraction pixels number, calculate
correspondence percent.
MIPPR 2003, Beijing
19
SRSIE
Experiment and Results

• Calculation

Error Extraction
Misses Extraction
Right Extraction
Missed Extraction
Right Ratio
Error Ratio
Error Extraction
Missed Ratio
MIPPR 2003, Beijing
20
SRSIE
Experiment and Results

• Results Analysis

Rarefied residential area distribution without
obvious road can obtain good result even the
narrow connection between two residential areas
Concentrated residential area distribution with
road come to good result for the road is totally
inside the residential area Smooth Bound for the
Gauss blur processing reason
MIPPR 2003, Beijing
21
SRSIE
Experiment and Results
Error
Error
MIPPR 2003, Beijing
22
SRSIE
Large range residential area
Aero photogrammetry image with scale of 115000
MIPPR 2003, Beijing
23
SRSIE
Small range residential area
Aero photogrammetry image with scale of 115000
MIPPR 2003, Beijing
24
SRSIE
SPOT-15m
Low resolution comes to good result while some
area inside residential area is a problem
MIPPR 2003, Beijing
25
SRSIE
Experiment and Results

• Statistic Results Analysis

(1) The right extraction ratios are all higher
than 90. (2) The average right percent is over
than 94.2
MIPPR 2003, Beijing
26
SRSIE

• Conclusions and Discussion

• Conclusion

(1) The semi-automatic extraction results of
residential areas from different types of high
resolution images show that the method is simple
and efficient and only takes very short time with
the 94.2 correctness to the residential area
extraction.
(2) The semi-automatic extraction is reasonable
for the current appication
(3) The residential areas extracted by this
method are available to provide information for
some application such as planning and
decision-making.
(4) The residential areas extracted by this
method can be used for the large scale mapping
and surveying with high resolution remote sensing
image.
MIPPR 2003, Beijing
27
SRSIE

• Discussion

(1) How to differ the road and road in the
residential area?
MIPPR 2003, Beijing
28
SRSIE

• Discussion

(2) Automatically Seed Selecting method?
Automatic Extraction
Semi-Automatic Extraction
MIPPR 2003, Beijing
29
SRSIE
Thank you!
Question?
MIPPR 2003, Beijing