138 
COMPRESSION OF IMAGE DATA FROM REMOTE SENSING SATELLITES 
J.Cl. Degavre 
European Space Research & Technology Centre 
Noordwijk, The Nether!ands 
Comission II 
Abstract 
A summary is presented of the results obtained by the European Space 
Research & Technology Centre (ESTEC) from its experiments of the 
compression on remote sensing satellites image data. Examples of the 
required instrumentation are given for both on-board satellite and ground 
applications. 
1. INTRODUCTION 
The successful utilization of image data from remote sensing satellites 
has stimulated the development of imaging sensors. The technology will 
satisfy the user's desire for finer spatial and spectral resolution. 
Forecasts suggest that 10m spatial resolution and 20nm spectral resolution 
design goals will be met in the next few years [1, 2]. The new data rate 
generated by such sensors will reach several gegabits per second and will 
exceed the planned data relay satellites capability. Efficient encoding, 
or data compression, in combination with the selection in flight of 
subsets of the imaging sensor capabilities will be mandatory. 
As part of its on-board satellite signal processing technology programme, 
the European Space Research and Technology Centre (ESTEC) has conducted 
several studies on the design of image data compression algorithms 
suitable for remote sensing image data. An algorithm has been selected 
and tested extensively on satellite imagery. 
2. | THE COMPRESSION ALGORITHM 
  
A compression algorithm has been developed at ESTEC by D. Chaturvedi [3] 
It consists in quantizing and encoding the image data after Cosine 
Transformation. The transformation, quantization and encoding are 
performed, block by block, over the complete image scan. The block is 
typically a square of 16x16 pixels. 
After Cosine Transformation, the data are much less correlated than the 
initial image grey levels and can be coded with less redundancy. Because 
natural images tend to have decreasing energy versus increasing spatial 
frequencies, the quantized data in the Cosine Domain can be pre-arranged 
in order of decreasing word length and coded with an efficient variable 
word length encoding scheme. The word length of each Cosine Transformed 
data sample depends of its amplitude and of the selected quantization 
step. The image is reconstructed by reversing all the operations after 
the word synchronisation has been recovered from the auxiliary data. 
The functional block diagram of the compressor/decompressor is shown on 
fig. 1. The compression ratio obtained varies from block to block and is 
a function of the image spatial frequency content. 
2.1 Theoretical Performance 
  
As the quantization of the Cosine Transformed image data samples is 
performed with uniform step and since the transformation is unitary, the 
mean squared difference between the origina! image and the image 
reconstructed after compression and decompression is given by the