004 
— 
zital 
2]. 
omp 
51- 
PC. 
the 
092. 
ital 
RS, 
1sed 
ton, 
ring 
best 
nical 
GIS AND CONTEXT BASED IMAGE ENHANCEMENT 
Manfred Ehlers * *, R. Welch®, Y. Ling h 
* Research Center for Geoinformatics & Remote Sensing FZG, D-49364 Vechta, Germany - mehlers@fzg.uni- 
vechta.de 
? Center for Remote Sensing and Mapping Science CRMS, Dept. of Geography, University of Georgia, Athens, GA 
30605, USA - (rwelch, yling)@uga.edu 
Commission IV, WG IV/6 
KEY WORDS: Visualization, Fusion, Automation, Segmentation, IKONOS, Quickbird, GIS 
ABSTRACT: 
Various applications require an optimized (and rapid) display of remote sensing imagery. Multisensor and multiband remote sensing 
data have to be enhanced for optimal band combination and image contrast. Problems that exist are that contrast stretches are usually 
optimized for the whole image and might not prove appropriate for selected features. For example, an image that contains land, 
water and beach classes will be stretched in a way that would produce a compromise for the different classes. Water is usually dark 
(especially in CIR display), beach will be very bright with little discernible structure (similar for urban classes), and other land 
classes (e.g. vegetation) will not make use of the full possible range of digital numbers. Also, different features might require 
different band combinations for optimum display. Selected stretching especially for regions of low contrast is nothing new in the 
analysis of remotely sensed data. Usually, this is done interactively by the analyst either by selecting a box or digitizing a certain 
area of interest in the image. This area is then enhanced using standard image processing techniques. The subset is then displayed 
separately to highlight certain features that would have been impossible to discern in a global enhancement mode. 
The goal of this study was to develop automated procedures for feature based image enhancement techniques. Feature based 
enhancement means that different feature classes in the image require different procedures for optimum display. The procedures do 
not only encompass locally varying enhancement techniques such as histogram equalization or contrast stretch but also the selection 
of different spectral bands. There are two main sources for this kind of information: (a) storage of a priori knowledge in a GIS, and 
(b) context based image information that can be extracted through a segmentation process. Both techniques can also be applied for 
optimum feature class selection. In this paper, we develop a five-step automated procedure for selective image enhancement and 
compare the results to those achieved by standard methods. The test area is a coastal region in the US which covers land, water and 
beach areas. Datasets are multispectral IKONOS and Quickbird satellite data. It is shown that the new methods produces superior 
results. 
1. INTRODUCTION water and beach classes. Using global image enhancement 
techniques, the image will be transformed in a way that would 
Image enhancement techniques are usually applied to remote produce a compromise for the different classes. Water is usually 
sensing data to improve the appearance of an image for human dark (especially in CIR display), beach will be very bright with 
visual analysis. Enhancement methods range from simple little discernible structure (similar for urban classes), and other 
contrast stretch techniques to filtering and {mage transforms land classes (e.g. vegetation) will not make use of the full 
(see, for example, Jensen 1996 or Gonzales and Woods 2002). ^ possible range of spectral values. Also, different features might 
Image enhancement techniques, although normally not required require different band combinations for optimum display. This 
for automated analysis techniques, have regained a significant cannot be done using conventional enhancement and display 
interest in recent years. Applications such as virtual strategies. Water, for example, may reveal more information in 
environments or battlefield simulations require specific RGB display whereas vegetation requires a CIR approach. The 
enhancement techniques to create ‘real life’ environments or to indicated problems will only get worse with the availability of 
process images in near real time. hyperspectral data where possible combinations of narrow 
bandwidth spectral channels can differ for land and water 
This paper deals with the development of automated techniques features. 
for the rapid enhancement of high resolution imagery from 
different sources. We will present the results for multisensor The proposed methods make use of existing GIS information, if 
and multiband remote sensing data that have to be optimized for available, and/or image preprocessing such as NDVI 
band combination and image contrast using standard point calculations. Using this approach, it is possible to design a 
operation functions such as linear contrast stretch or histogram procedure for completely automated image enhancement that 
equalization techniques. Problems with these rapid works in an optimized way for the selected features. 
enhancement techniques are that they are usually optimized for 
whole images and might not prove appropriate for selected 
features. This affects especially coastal areas that contain land, 
  
Corresponding author 
397