de Carvalho, Luis 
  
CHANGES AT MULTIPLE SPATIAL SCALES 
Luis M. T. CARVALHO, Leila M. G. FONSECA", Fionn MURTAGH", Jan G. P. W. CLEVERS**** 
"Wageningen University, The Netherlands 
Centre for Geo-Information 
Luis.carvalho@staff.girs.wag-ur.nl 
"National Institute for Space Research, Brazil 
Division of Image Processing 
LeilaQ dpi.inpe.br 
"Queen's University of Belfast, Northern Ireland 
School of Computer Science 
F.murtagh@ qub.ac.uk 
"Wageningen University, The Netherlands 
Centre for Geo-Information 
Jan.clevers @staff. girs. wag-ur.nl 
Technical Commission Session IV-5 
KEY WORDS: Change Detection, Data Fusion, Forestry, Image Processing, Multi-scale Database, Wavelet 
Transform. 
ABSTRACT 
Change detection on rasterized data is extremely dependent on accurate radiometric and geometric rectification. The 
development of processing tools able to minimise these requirements has been recognised since the late eighties. In the 
present paper we present a methodology for detecting changes on multidate satellite images with different radiometric 
and geometric characteristics via Multiresolution Wavelet Analysis. An area in south-eastern Brazil was chosen as case 
study. In the last 20 years the site was characterised by an increase of mining activities and deforestation. Landsat TM 
and MSS images from July 1981, November 1985 and August 1998 were used. The idea is to decompose a set of 
images into averages (overall pattern) and details images at different resolutions. Image differences due to the effects of 
spatial misregistration, atmospheric condition and sensor characteristics are depicted across scales. No radiometric 
rectification was applied to the input images and the spatial misregistration ranged from one to three pixels. To detect 
deforestation and new mining areas we used details at the third and fourth scales. Deforested areas as well as new 
mining sites were successfully pinpointed without previous radiometric rectification or threshold definition while 
differences not related to land cover changes were bypassed. Misregistration effects and small area changes are depicted 
as fine details. Phenological characteristics, atmospheric effects and differences in sensor calibration are represented at 
coarser scale levels. Hence, using information from intermediate scale levels one can minimise the problems mentioned 
above. 
1 INTRODUCTION 
Multitemporal and multisensor analyses of satellite images are becoming important research fields in geo-information 
sciences. This is mainly due to large quantities of remotely sensed data accumulated over the last twenty-five years. We 
are also waiting for even more information to come from a great number of new and powerful Earth observation 
systems. Improvements on spatial and spectral resolution mean more gigabytes for a computer to manipulate. Hence, in 
order to explore all these sources of information, effective ways of fusing, analysing and storing this huge amount of 
data must be developed. 
Digital change detection is also closely related to the issues mentioned above (Wong et al. 1997). For example, the 
combination of early images from Landsat MSS with upcoming data from Ikonos must be possible if we want to get the 
most from historical and up to date information. Analyses of dynamic processes with such a data set would provide 
useful inputs for historical characterisations, modelling and decision-making. The techniques available nowadays for 
detecting changes on rasterized data are extremely dependent on accurate radiometric and geometric rectification (Dai 
and Khorram 1998, Schott et al. 1988), which are very difficult tasks in some situations (e.g. poor quality of old 
sensors). The development of automatic analysis tools that could be able to minimise these requirements is recognised 
since the late eighties by Singh' s classical review on change detection (Singh 1989). 
  
340 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000.