METHODS AND TECHNIQUES FOR THE CORRECTION
OF NATURAL SHADES ON AERIAL-PHOTO OR SATELLITE MAPS
Nikolaos Tzelepis and Byron Nakos
Cartography Laboratory, Department of Rural and Surveying Engineering
National Technical University of Athens
9, Heroon Polytechniou STR., Zographos, GR-157 80, Greece
email@example.com - firstname.lastname@example.org
This study examines possible ways first to restrain or even eliminate the “false” shades of the relief,
created mainly by the direction of the natural sunlight presented on aerial-photo or satellite images, and
then to replace them with the “correct” artificial hill-shading shades according to the cartographic
principles of legibility and perceptibility, which allow the map user to easily and clearly interpret the
relief’s shapes and formations. The natural shades are eliminated from the aerial-photo or satellite
images by applying special radiometric and statistical processing in order to create images free of
shades. An analytical description of the earth’s surface -stored as a digital elevation model- combined
with computing tools of a variety of hill-shading methods are used to produce a cartographic “correct”
image of shades of the study area. Finally, image-processing techniques are applied to create new images
composed by the “correct" shades and the shaded free aerial-photo or satellite images. The outcomes
indicate the potential of the incorporated methods and applied techniques in order to construct aerial-
photo or satellite maps clearly and legibly perceived.
KEY-WORDS: Aerial-photo/satellite maps, relief visualization, hill-shading, illumination models,
correction of topographic effect.
The established map definitions are generally based on the graphical representation of the real world and
the intrinsic phenomena or spatial relationships, where images and photographs are not conventionally
parts of graphical symbolization. However, the various definitions introduced through the time, show that
important changes have been accomplished in all mapping tasks and procedures, -mainly due to
technological development- and lead to re-examination of the effectiveness of existing cartographic rules
and principles [Nakos and Filippakopoulou 1993]. In some of the proposed definitions, the content and
use of map are related to a wider range of activities, involving any graphic image that may include a
spatial component, and subscribing to cartography’s apparent interest for new products [Kraak 1989].
In most of the cases, images of the earth’s surface have been used for cartographic purposes in
an indirect way, as sources of collecting data through photogrammetric and remote sensing techniques.
Late advances in the use of aerial-photo or satellite images have come to add new capabilities. These are:
- The continuous development of the earth observation systems and sensors used for data acquisition, by
the means of increasing resolution.
- The increasing availability of the data produced by these systems even to the simple users.
- The consequent release of the appropriate software that supports the tools to take advantage of the new
The potential of these new capabilities enable users involved in geo-informatics to easily apply digital
mapping procedures and produce graphical compositions combining aerial-photo or satellite images with
existing vector cartographic data. This utilization of earth’s images can offer a strong cue of familiarity to
map users, stimulating their cognitive approach to support cartographic perception. Map design and
production may be challenged by the exploration of the potential adjusting theoretical cartographic
principles on the complex background of aerial-photo and satellite mapping.
One of the most critical components of these new cartographic applications is the representation
of the topographic relief, which is already an inborn characteristic of the image, caused by the natural
shades of the relief created from the incident sunlight. In several circumstances, the produced realistic
shades make the earth’s topography inversely perceived; whether the natural sunlight is directed east or
even southeast. Cartographic methods of hill-shading use analytical techniques of computing relief
shaded images characterized by a clear and legible perception of the earth’s morphology. Such shadings
could be added as a supplement to the original images [Yoeli 1965]. However the visual result will be
improved if the share of relief’s influence is isolated and removed from the existent tones, generated not
only from the relief but also from various other factors as the vegetation, the earth texture, the weather,
the conditions of sunlight, etc.
Elimination of the natural shades in aerial-photo and satellite images
The natural shades of the relief present in aerial-photo and satellite images are not always created with
proper illumination geometry such as the basic rules of cartographic hill-shading suggest, in order to
achieve a correct interpretation of the topography. In fact, the technical schedule of satellites’ capturing
procedure is usually made to ensure that a clear image of the natural surface of the earth will be acquired,
and this means that it occurs early in the morning when the sun comes from the east. Aerial-photos’
capturing is easier to be programmed, and it usually occurs closer to the noon to avoid heavy cast
shadows that conceal surface information. However, in the above illuminating conditions still some
shading does occur, and in addition there are exceptions from the time schedule which cause intense
shading. So, if the sun is low enough, then the “fault” shading and shadows are respectively heavy and
they must first be omitted before applying the computed “correct” ones. Shading is a part of the entire
physical process of illumination and radiance recording, on which remote-sensed data capturing is based.
More specifically, the recorded radiance values of earth’s surface could ideally correspond to the
objective quantities of light reflected by the ground, depending only on the energy characteristics of the
land-cover. In practice, there are two main reasons affecting the final recorded values: the atmospheric
attenuation caused by scattering and absorption, and the geometric effect of topography on ground
reflectance process [Krauss 1991, Jensen 1996]. The change in radiance due to atmospheric attenuation is
a very complex phenomenon, that several alternative estimations have been suggested for its description.
Although its influence on the whole process differs from the case of satellite to air-born capturing, in both
cases its visual impact works in a global way. In the context of the present study, attention is focused on
the topographic effect, which is the main reason for the formation of the locally dependent visual
impressions of shading and shadows.
Topographic normalization is a basic radiometric correction, occurring in the preprocessing of
remotely sensed images, and its calculation is carried out by extensive studies. The proposed correction
methods refer to either an integrated model for atmospheric corrections, definitely requiring ground
control measurements and knowledge of observing conditions’ details, or to more practical mathematical
and/or empirical models based exclusively on the topography of the surface [Karathanassi et al. 2000].
Below are presented two commonly used mathematical methods, the rather simple cosine correction and
a more effective one called the Minnaert correction:
LH = corrected radiance, corresponding to a horizontal surface
LT = recorded radiance, corresponding to sloped terrain
sz = sun zenith angle
i = incidence angle between sunlight and surface normal
k = Minnaert constant, varying between 0 and 1
As it is obvious from the latter equation for k=1 the Minnaert correction degenerates to cosine correction.
Evaluations of the known mathematical or empirical methods provide important information for their
effectiveness [Meyer et al. 1993]. The cosine correction, which is exclusively based on the assumption of
a Lambertian reflection of an isotropic distributed light by the illuminated surface, tends to overestimate
shading in rugged terrain with weakly illuminated areas, thus it is mainly applied in flat terrain. On the
other hand, the Minnaert correction takes also into account the component of light caused from diffusion
and surrounding reflection, by applying a constant that implies the extent to which a surface is
Lambertian. Furthermore, only small difference in results is observed between the Minnaert correction
and other complex statistical methods present in literature. Thus, in the present study the above two
methods were used for practical reasons.
In this work, two samples of earth’s surface images, one aerial-orthophoto and one satellite
image, are used for an experimental procedure of shading correction as large-scale or small-scale
representations respectively (Figure 1). The angles of sunlight for the satellite sample were found by the
metadata information of older scenes of the same date, since the scene used came without this
information. For the aerial-photo sample, the time and date of the shot were used as input data to calculate
the polar coordinates of the sun position. The respective digital elevation models were interpolated using
collected height information, by means of either digitized hypsometric contours and breaklines from maps
of scale 1:50,000 (case of satellite data), or individual spot heights captured from stereo-pair images (case
of the aerial-photograph). At this point a critical issue arises referring to the matching between the digital
elevation model and the image to be corrected.
AT TRIKALA, CENTRAL
AREA: 8.1X7.3 KM2
ELEV: 141-880 M
CELL SIZE = 20M
SUN AZIM = 152°SE
SUN HEIGHT = 60°
SAMPLE OF AERIAL-
(AREA WITH QUARRIES AT
NORTHWEST OF ATHENS)
AREA: 240X220 M2
ELEV: 828-923 M
CELL SIZE = 0.4M
SUN AZIM = 140°SE
SUN HEIGHT = 68°
(a) (b) (c)
Figure 1: Sample imagery and relative available information for the corrections
(a) 3D visualizations of the samples’ areas with draping of hypsometric tints over the interpolated DEM
(b) The respective samples of satellite and aerial-photo imagery
(c) Computed shading images corresponding to the position of the sun at the date and time of data capturing
Both methods of correcting topographic effects are applied on the two samples, where the Minnaert
constant (k) is approached by successive approximations of values between from 0 and 1. The adopted
value of k was empirically estimated by selecting visually the most effective result. The final image
values (LH) calculated by multiplying the corrections with the existent values (LT), are normalized to the
range of grayscale visualization. Evaluating visually the results of the computations applied on the
satellite image sample (Figures 2a,b), it appears that the cosine correction indeed causes overestimations.
Contrary, the Minnaert formula applied with a constant valued equal to 0.6, gives satisfactory removal of
shades with the consequent loss of three-dimensional impression.
The case of the aerial-photo (Figures 2c,d) has certain difficulties. First, the relief corresponding
to images of large scale is very detailed and complex. The slopes at locations with rough terrain are larger
that the expected physical ones, resulting in extremely high values of corrections at the adequate
locations. Assuming that the maximum logical slope on the ground is up to 50%, the maximum correction
is computed for the relief of this area, and values larger than this are ignored. The aerial-photo’s
corrections then, become similar to the ones computed for the satellite sample. In this way, the problem of
overestimation using cosine correction is not apparent and this method produces more satisfactory results
than Minnaert’s. An even more inconvenient problem is that the shading tones interfere with tones of
different reflection due to land-cover, whose variation is certainly higher. This lack of three-dimensional
impression of natural shading in aerial-photos, due to the covering and flecked texture of earth’s surface
had been identified long ago, while browsing ways to model the early efforts on manual skillful hill-
shading [Imhof 1982]. Still, it is clear to be seen that shades of relief have been omitted, if the corrected
image is closely examined together with the simulated image of the relief shading according to the date
and time of capturing.
(a) (b) (c) (d)
Figure 2: Grayscale visualizations of corrections (“the larger the lighter”) and respective results
(a) Cosine corrections on the satellite sample
(b) Minnaert corrections (k=0.6) on the satellite sample
(c) Cosine corrections on the aerial-photo sample
(d) Minnaert corrections (k=0.5) on the aerial-photo sample
Relief-shading for combined visualization with other surficial representations
The role of relief representation as a map component, varies from being a supplementary background that
enhances the depiction of the main portrayed phenomena [Wheate 1996] to “the foundation for all the
remaining contents of the map” [Imhof 1982, p. v], depending on the purpose and type of the map. In any
case, the continuous three-dimensional nature of relief makes the choice of its symbolization a
particularly complicated and special-treated map procedure [Robinson et al. 1995]. One of the earliest
developed methods for relief representation is the method of hill-shading, which gives a realistic and
easily perceived depiction of the relief even for casual users. However, cartographers had to wait for
contemporary technological achievements in order to overcome the difficulties in production, and
efficiently practice and experiment with hill-shading images.
The cartographic method of relief-shading or hill-shading is the visual differentiation of tone
under specific light conditions and due to the variable local orientation on each point of earth’s surface,
counted with the percentages of the reflected light. In reality the interplay of illumination with a surface is
expressed using many parameters which denote more complicated physical models and position-
dependent effects of illumination, like cast shadows, illumination from neighbor surface points or
atmospheric perspective. But as cartographic hill-shading is used for representing and not simulating
reality, effects like these are not implemented because they do not necessarily serve the perception of the
surface shapes [Imhof 1982, Horn 1982]. Instead of that, convenient assumptions that ensure legible
shading images are considered; Lambertian, perfectly diffusing surfaces that demand relatively simple
calculations, and hypothetical sunlight that psychological reasons enforce its placement to west directions
are applied. Further guidelines are the fitting of main light direction to the dominant aspects and slopes of
relief, and adjustment of light source according to local surface orientation. Among the several methods
of representing terrain information, hill-shading applies for its ability to present a quick, comprehensive,
non-skilled and familiar image of topography. The aforementioned characteristics make it the most proper
method to visualize terrain information in combination with other phenomena that are also portrayed by
using surficial symbolization, eg. polygonal thematic data, discrete-zoned or continuous realistic
hypsometric tints, conventional analogue maps which have been scanned and geo-referenced, or images
of earth’s surface.
The production of a relief-shading layer for combined surficial representation has two main
phases of processing, where certain choices can define the final result. First and foremost, is the main
production of the hill-shading tones, requiring a digital elevation model of the area, and the selection of
the most convenient available shading model and the appropriate parameters. The common parameters for
any model are the azimuth and zenith angle of the hypothetical sunlight direction –usually placed to
north-east and 50% from the ground- while special other parameters for a certain model might also exist.
It is very important to note here that, despite of the significant progress of research about hill-shading,
only a restricted range of certain models can be found in the computational environments of the
commercial and staple software products used for cartography. If a wider choice of selection is needed,
either use of other type of software must be developed (eg. based on 3D computer graphics) or
programming of routines should be implemented along with the necessary tools for data-exchange.
According to the type of combined spatial representation, some general –even mostly empirical so far–
aspects of choosing the appropriate kind of model may be applied, but still it is only the visualization of
shading that can affirm if the choice of parameters of shading is adequate or not.
Secondarily, the produced grayscale image of shading is further adjusted towards its tonal
characteristics and texture, according to the combined spatial representation. The basic concept of this
fine-tuning is to confine the shading values to a subset of the full range covering from black to white. The
adjustments that can be made consist of the improvement of brightness and contrast, and the smoothing of
imperfections or unnecessary details using appropriate filters. At large scales shading aims to emphasize
the dominant forms of the topography while at smaller scales more detailed information is needed, and
the maximum intensity is limited by the relationship between the shading and the legibility of map
[Keates 1989]. Successful combination of the component representations is achieved when their features
as they are presented in the resulted image are supported by the balanced interplay of each component’s
As these modifications of hill-shading tones depend totally on the combined surficial
representation, they should be accomplished together with the final implementation of two respective
images. During the procedure of combination, there is one more way to optimize the visual result. By
using a regulatory factor a, there can be a balance between the fully mixed image and the initial one
[Tzelepis 2000] as described by the following equation:
This is a characteristic way of image combination, available in most of the image-processing software.
This feature together with the ability of direct previewing, are very helpful tools for the successive
implementation of a hill-shading method.
Combination of computed hill-shading and a non-shaded image of earth’s surface
A successful incorporation of relief-shading representation in a satellite or an aerial-photo image free of
shades is achieved when the critical features of the two components are preserved, or even further
enhanced. A primal approach to this effort is an understanding of the particularities, carried out by the
close examination of types and functioning of the visual elements for each component. As indicated by a
conceptual framework of image analysis, tone or color is a fundamental property of imagery, while other
visual elements like shape, pattern or texture are expressed by means of spatial arrangements of it,
sequentially perceived in an intermediate or higher level [Estes et al. 1983]. All elements are functioning
in the context of several image analysis tasks (detection, identification, measurement, knowledge-assisted
labeling and significance), for aiming at the interpretation and perception of the spatial objects and at
relationships among them. In hill-shading images a similar situation is evolved; tone is the fundamental
property, shapes and patterns are formed by arrangements of tone, revealing the forms of relief in the
presented area. The visual element of size is also apparent in both components, but it functions with more
detail in the context of the imagery. Generally, these two visualizations seem to be perceived in a similar
way, subscribing to their combination for producing harmonic visual result –here carried out by
multiplication of the respective images’ values at corresponding points.
(a) (b) (c) (d)
Figure 3: Computed hill-shading images and combinations with the satellite imagery sample.
(a) Standard model of Lambertian surface based on the perfect diffusion assumption. The sunlight azimuth
selected for better shading is 156° NE.
(b) Local adjustment of azimuth and elevation of light direction for better lightning of the area, relatively to the
result of 3a (here equal to 25% for each angle).
(c) Fine-tuning of shading on example 3b, by tonal modifications (increased brightness 50% and contrast 25%).
(d) An alternative choice with similarly balanced results, produced by a simple mathematical model (parameters to
be defined are the gray tone for horizontal surfaces and the rate of change for the gray tone to slope ratio, here
equal to 50% each, brightness and contrast of shading image have been improved).
The presence of hill-shading adds to the image the essential cue for depth perception, which is required in
order to operate as a non-perspective simulation of three-dimensions [MacEachren 1995]. In contrary to
the original captured images, the relief forms perceived using proper shading rules correspond to the
actual ones. Moreover, for those spatial objects that are shaped along with underlying relief formations, it
can be seen that perception is impressively enhanced. In large-scale images (aerial-photos or high
resolution satellite imagery) objects like these can be artificial features like road network or modulations
around buildings. In this case of course these objects must be also included in the utilized digital terrain
model. The beneficial influence of representing the relief with correct hill-shading can be seen in all
stages of image analysis; from the early stage of detection where mental images are recalled to help the
observer to label the object, to more time-consuming consideration needed for more complex
arrangements, where special related knowledge might be required. In the contrary, there are cases where
the perception of relief is helped by locative information derived from the image, recalling existent
special knowledge related to the site. In fact, the interpretation of objects and spatial relationships
presented in the images of earth’s surface, and the perception of forms of relief accentuated under
conditions of correct shading, are mutually enhanced.
The corrected samples utilized for this work, are combined with hill-shading images computed
from the staple model of Lambertian reflection (Figures 3a, 4a) or other alternative methods. More
specifically two methods are tested for the satellite image; the adjustment of light source for better local
reflection and brighter shades (Figure 3b) and a simple mathematical method based on balancing the
result around a defined gray tone for horizontal surfaces (Figure 3d). For the aerial-photo the same model
was applied, assigning the horizontal surfaces with a darker tone (Figure 4b). An example is given also
using a method of specular reflection designed for use on computer graphics, which gives an image of
“plastic” shading that fits the realistic impression caused by observing in a close distance (Figure 4d). The
dominant aspects of relief are found, indicating convenient directions for hypothetical light sources.
Further modifications of brightness and contrast are made to optimize the shading towards to incorporate
it into the imagery (Figures 3c, 4c). The tuning of the shading image in order to reveal the relief forms
consists of delicate and critical graphical decisions taken under serious constraints. The fact, that slight
changes of shading are impossible to be seen in the continuous presented information of the images,
suggests that enforcement of contrast should be high enough to permit the detection of relief shapes. On
the other hand special care is needed to avoid the presence of dark tones disturbing the legibility and
readability of imagery’s information, which is still the first priority of this combination.
(a) (b) (c) (d)
Figure 4: Computed hill-shading images and combinations with the aerial-photo imagery sample.
(a) Standard model of Lambertian surface based on the perfect diffusion assumption
(b) Utilization of a mathematical model for more balanced results, because of the dark tones caused by existence
of steep slopes (same parameters to be defined as in [Figure 3d], with the gray tone for horizontal surfaces here
equal to 12.5% and the rate of change for the gray tone to slope ratio, equal to 50%)
(c) Fine-tuning of previous shading by tonal modifications (with increased brightness % and contrast 25%)
(d) A different choice produced here by a mix of 50% Lambertian and 50% specular reflection for computer
graphics (brightness and contrast of shading image have been improved)
At the context of this work, an attempt has been made to experiment with optimal visualizations of the
relief-shading in the new fast-spreading satellite and aerial-photo mapping products. For practical reasons
of simplifying the procedure at this first approach, only grayscale images were used. The several tasks of
the procedure were carried out using commercial software for GIS and conventional image processing
(ARC/Info, Adobe PhotoShop) and custom programming routines. First, existent natural shades that
mystify the perception of users were removed using techniques of radiometric pre-processing. The
calculation of corrections from direct reflection by these techniques, based on the Lambertian reflection
law, is only a part of an integrated procedure for environmental noise. The inclusion of more parameters
like reflection from neighbor areas or scattering can be further examined for more reliable computation of
The complexity of the imagery’s information itself requires that the production of the hill-
shading representation needs to be implemented along with the final task of its incorporation in the image,
in order to avoid disturbances on image reading and final product legibility. Strong visual tools like direct
previewing are essential for getting out the most of this delicate graphical procedure. This need becomes
even more important if color images are to be used, introducing another critical point of the selection
among alternative combining procedures based on different color models or graphical techniques. The
efforts described herein constitute some general experimentation based on the characteristics of the single
component visualizations. For an objective documentation on the perceptibility of the combinations of
hill-shading methods and the images, it is suggested that related questionnaire including visual examples
like the above, should be planned and applied on users, so that data for statistical processing to be
provided. In addition, it is a strong challenge to elaborate these first results in order to develop the entire
project to three-dimensional visualization.
Estes, J.E., Hajic, E.J., and Tinney, L.R., 1983. “Fundamentals of Image Analysis”. In Manual of Remote
Sensing, (Coldwell, R.N., Ed., Vol. 1). American Society of Photogrammetry, Sheridan Press, New
Horn, B.K.P., 1982. “Hill shading and the reflectance map”. Geo-Processing, 2: 65-146.
Imhof, E., 1982. Cartographic Relief Presentation, (Steward, H.J., Ed.). Walter de Gruyter, Berlin.
Jensen, J.R., 1996. Introductory Digital Image Processing, (2nd ed.). Prentice Hall, Englewood Cliffs.
Karathanassi, V., Andronis, V., and Rokos, D., 2000. “Evaluation of the topographic normalization
methods for a Mediterranean forest area”. Proceedings of International Archives of Photogrammetry
and Remote Sensing, 33(B7): 654-661.
Keates, J.S., 1989. Cartographic Design and Production, (2nd ed.). Longman Inc., New York.
Kraak, M., 1989. “Computer-assisted cartographical 3D imaging techniques”. In Three dimensional
applications in Geographical Information System, (Raper, J., Ed.). Taylor & Francis, London: 99-
Kraus, K., 1991. Photogrammetry, (Vozikis, E., Ed.). Technical Chamber of Greece, Athens. (Translated
MacEachren, A.M., 1995. How Maps Work. The Guilford Press, New York.
Meyer, P., Itten, K.I., Kellenberger, K.J., Sandmeier, S., and Sandmeier, R., 1993. “Radiometric
corrections of topographically induced effects on Landsat TM data in an alpine environment”.
Photogrammetric Engineering and Remote Sensing, 48(4): 17-28.
Nakos, B., and Filippakopoulou, V., 1993. General Cartography. Lecture Notes, Department of Rural
and Surveying Engineering, National Technical University of Athens. (In Greek)
Robinson, A.H., Morrison, J.L., Muehrcke, P.C., Kimerling, A.J., and Guptill, S.C., 1995. Elements of
Cartography, (6th Ed.). John Wiley & Sons, New York.
Tzelepis, N., 2000. “Digital Image Processing Techniques for Combined Representation of Relief
Shading in Maps”. In Proceedings of 6th National Cartographic Conference. Hellenic Cartographic
Society, Athens. (In Greek)
Wheate, R., 1996. “Re-examining the Cartographic Depiction of Topography”. In Cartographic design:
Theoretical and practical perspectives, (Wood, C.H., and Keller, C.P., Eds.). John Wiley & Sons,
Yoeli, P., 1965. “Analytical Hill Shading”. Surveying and Mapping, 25(4): 573-579.