Jos

1
GEOINFO 2004 VI Brazilian Symposium on
GeoInformatics
AUTOMATIC GEOMETRIC AND RADIOMETRIC REGISTRATION
OF LANDSAT-TM IMAGES USING MUTUAL INFORMATION

• José Pinheiro de Queiroz Neto - CEFET/AM
• Mário Fernando Montenegro Campos - UFMG
• Bruce Walker Nelson - INPA
• José Luiz de Souza Pio - UFAM

2
The Problem

• IMAGE REGISTRATION
• Find correspondence points between two or more
  images of same orbital sensor acquired at
  distinct times, considering
• Geometric distortions, and
• Radiometric distortions.

3
The Problem

• Geometric Distortions.
• Caused by position, size and orientation of a
  pixel altered in acquisition process.
• Radiometric Distortions.
• Affect pixel encoded radiance

4
231_062 scene - bands 3,4 and 5
1999
1995
5
Motivation

• Studies requiring change detection between images
  acquired at distinct times have been a difficult
  task for researchers
• Without registration, programs to produce and
  analyze data from the same region will have
  inaccurate results.

6
Goal

• This work aims at studying and implementing a
  methodology to automate registration of
  multi-temporal images, correcting both geometric
  and radiometric distortions.
• ARMI Automatic Registration based on Mutual
  Information

7
Concepts

• Mutual Information
• Geometric and Radiometric Transformation
• Optimization Method (Powell Method).

8
Mutual Information

• Statistical technical based on entropy
• A similarity measurement

9
Geometric Transformation

• Affine Transformation

10
Radiometric Transformation

• Irradiance
• Ltot (?ET/p) Lp
• L K1?K2
• Digital Number (DN)
• DNK3LK4
• DNm?b
• Relation between DNS and DNR
• DNS(mR/mS)DNRbR- (mR/mS)bs
• G(DNR) ga DNR of

11
Radiometric Transformation

• A linear transformation G ga.DNof
• Apply a gain (ga) in pixel value
• Add an offset (of) in pixel value

12
Optimization Method (Powell)

• a arg mina(-IMRS(a))
• Search for the minima of N dimensions of MI
  through successive minimization, one dimension at
  time through a set of M different directions.

13
ARMI Fluxogram
14
Geometric results
15
Geometric results
16
Geometric and Radiometric results
17
Geometric and Radiometric results
18
1995 x 1985 - before
1985
1995
19
1995 x 1985 - after
1985
1995
20
1995 x 1988 - before
1988
1995
21
1995 x 1988 - after
1988
1995
22
1995 x 1990 - before
1990
1995
23
1995 x 1990 - after
1990
1995
24
1995 x 1999 - before
1999
1995
25
1995 x 1999 - after
1999
1995
26
Real Case Ducke Reserve
27
1985
28
1988
29
1990
30
1995
31
1999
32
Conclusions

• Automatic registration method for satellite
  images time series which is able to
  simultaneously correct both the geometric and
  radiometric distortions without correspondence
  points.
• Limitations of the method
• high computational cost and the need for minimum
  entropy between reference and subject images.
• Future works
• introduce a multi-resolution technique to
  decrease computational cost.
• Integration with a GIS.

33
Questions?