|
Concurrently, current research into computational vision shows a need for software which allows one to collect various perceptual features and their characteristics as those work for someone watching a natural scene.
Whatever the scale of the study, large collections of photos are gathered (hundreds to thousands) : their analysis and their restitution pose some problems that computer science and multimedia can resolve.
To pass from the rough corpus of photos to a bank of images analysed and indexed, we are constructing an integrated computational tool.
It has two potential uses :
- first, to restore to the landscape its spatial continuum by the possibility
of managing some mosaics of images, spatialised and angularly oriented
;
- second, to manage some multitemporal series of landscape images and
to work on the different temporalities of the landscape.
Through a landscape study realised in the Arctic area (Svalbard, King's
Bay, 79°), we will present the architecture of the integrated tool
of landscape analysis, focusing particularly on three aspects :
- the tool of data acquisition for the analysis of the images ;
- the structure of the spatialised image bank ;
- the multitemporal approach through an automatic analysis of light
and colour.
To pass from the rough corpus of photos to a bank of images analysed
and indexed, we are constructing an integrated computational tool. It has
two potential uses :
– first, to restore to the landscape its spatial continuum by the possibility
of managing some mosaics of images, spatialised and angularly oriented
;
– second, to manage some multitemporal series of landscape images and
to work on the different temporalities of the landscape.
 |
The demonstration rests on a program of arctic landscape
research reali- sed in theKing’s Bay, 79° North, Svalbard, in 1998.
The study relates on the Basin of the Loven East Glacier (10 km2), from the fjord to the alimentation circus. |
Basin of the Loven East, 79° North, Svalbard
Basin of the Loven East, 79° North, Svalbard
The landscape can be captured through photographic sampling and a standardised grid of surveys. The photo-samples allow us to make an inventory of a physical state of the landscape. Processes of sampling of the area are developed to collect the photographs. The location of each camera shot is determined in a systematic way. The shots are organised along 7 transects. The density of the points vary with the characteristics of the spatial entity.103 points are set on the transects. The North-South distance between two points is 250 m on the sandur, 125 m in the moraine and at the front of the glacier, 250 m on the glacier, along 3 transects only, 500 m on the top of the glacier.
In each point, a principal of angular sampling is applied : 3 photographs of 45° account for the 360° of the landscape scene.
In each point, a random drawing determines the direction of the first camera shot between 0 and 120° from the North. The direction of the second shot is at 120° from the first one. The third direction is at 240° from the first one. The random drawing is re-done in each point of the basin.
 |
In each point,three horizontal photographs have been taken at a height
of 1,60 m.
A forth shot has been taken vertically in the direction of the ground. |
The four camera shots of each point constitute a bank of 400 images.
By a cartographic interface, one can reach each shot of each point.
Basin of the Loven East, 79° North, Svalbard
 |
|
Basin of the Loven East, 79° North, Svalbard
 |
 |
The computer power allows us to display an important number of images
and to relate to their geoposition.
Basin of the Loven East, 79° North, Svalbard
 |
 |
Basin of the Loven East, 79° North, Svalbard
 |
|
The landscape is not only spatially variable, it is also variable in time. The sampling at different time-steps allows us to account for the temporal variations of the landscape at different scales (year, season, day, hour).
Along the central transect, 14 points have been the object of 3 observational
surveys in 50 days.
 |
 |
The seasonal evolution is surveyed day after day in order to account
for the temporal continuum. During 50 days, from the nearest point to the
French Camp, 3 photographs were taken, each morning at 10h00 : the first
in the direction of the King’s Glacier, the second in the direction of
the Loven East, the third in the direction of Cape Mitra.
 |
 |
These three camera shots have been taken from the point 28, each morning
at 10h00 during 50 days from August 24th to October 9th 1998. Conversely,
the 25th of August and the 14th of September, 3 photographs have been taken
from the point 28 each two hours during 24 hours.
The poly-system “Landscape”
This diagram juxtaposes three boxes in interaction :
– the “productive system” : it is the “fabric” of the landscape, constituted
by the abiotic, biotic and anthropic components ;
– the “visible landscape” : it corresponds to the landscape scenes
as seen from the ground ; it has to be studied by itself as an architectural
pattern. It is composed of objects (hill, forest, village ...) and elements
of the image (forms, plans, lines, colours, light ...).
It is not possible to study the “visible landscape” directly through the “productive system” : this is why there is a “non reducibility zone” between these two boxes.
– the third box of the poly-system landscape is the “user system” :
the landscape is the object of consummation, scientific research and political
decisions.
– between the “visible landscape” and the “user system” there is a
filter which integrates the phenomenons of perception and interpretation
proper to a group and to the individual, without whom the landscape would
not exist.
The “user system” retro-acts principally on the “productive system”
(deforestation, urbanisation ...) and secondarily on the “visible landscape”
(landscape management). The different components of this poly-system “landscape”
can be scientifically studied. In this presentation we are simply going
to present our approach to the “visible landscape”, the central box of
the poly-system.
A computerised tool for data acquisition and treatments facilitates
the analysis of the landscape photographs. Some operations, like measurements
of surfaces, are automated.
User interface
• Scene box : Image select and display.
• Component identification box : Image features (objects or image elements)
drawing and viewing : positioning and contouring.
• Objects box : Description (filling and viewing) of an object of the
scene, or of a group of objects from an area.
• Image elements box - Description of image areas (filling and viewing),
that is characteristics of the surfaces and of the contours. The "lines"
section focuses on the apparent tridimensional shape of the scene.
• Composition and layers viewing
– Composition : For a drawed area or a selected feature,
viewing of its proportion in the image (and of related features.
– Layers : Viewing of all or some objects or image elements.
The treatment of the images of lansdcape allows us to establish typologies
of the landscape visible from each point of the studied area.
For example, the basic elements of the landscape of the King’s Bay are
listed on each photo and totalled by directional sector as for the total
of the 3 camera shots.
In another way, each landscape is a complex composition which varies
according to the point and the direction of the observation.
In our example, the 6 basic elements taken into account for the Loven
East Basin offer 41 possible combinations. 8 combinations are particularly
represented within the whole photos and constitute the matter of the typology.
This typology can be realised by directional sector or on the whole photos.
Each type can be illustrated by a representative photo.
Types of landscape combinations for the whole photos.
 |
 |
If one considers the landscape combinations of each photo, direction
by direction, one obtains a map of the combinations visible in each point,
in the three directional sectors.
   |
 |
 |
|
One can also ask oneself from which point is visible such or such basic
element of the landscape ?
From which point and in which direction can we see the ... ?
   |
Walking about in the landscape system of the Loven East, 79° N, Svalbard
 |
 |
 |
Concerning temporal dynamics of the landscape, three temporal scales can be considered :
– yearly cycles (local and global change),
– seasonal cycles (landscape phenology),
– daily cycles (linked to the apparent movement of the sun and to the
instantaneous climatic “ambiences”.
The day after day survey accounts for the temporal continuum.
To access to the bank of temporal images – taken from the nearest point
to the camp – one can choose the orientation, the time-step and the starting
day. For example, we choose to show one photo in the 320° direction,
each two days during 50 days, starting the 24th of August.
 |
The landscape changes hour by hour. The polar summer permits some continuous
survey for 24 hours. Here, the hour is the unit of the time-step.
|
Colours existing in the image are figured in this diagram.
Here are considered the only two dimensions of hue and saturation that
are represented through their colour and their position (respectively
for the latter : angle and distance to the centre). The luminance of colour-points
is related to the number of points with such a colour in the image. The
cross situates the mean value of the considered area. The centre corresponds
to grey and white, that is to totally desaturated colours.
The mean colour of an image is of course not enough to account for the
light “ambience” of a landscape. The sharing of colours in the image depends
on the physical location’s components that change according to hours and
seasons.
It is then necessary to consider several areas of the image that could
correspond to objects or features having some meaning.
It is then necessary to consider several areas of the image that could correspond to objects or features having some meaning.
One could have seen that the nature of clusters corresponding to colours of several areas of the image differed through quite a few aspects. Their dynamics appear in a similar way : their positions, orientations, patterns and dispersal progress according to various behaviour.
Applied to photos of a single landscape scene, taken at different moments
(seasons, days, hours), these automatic measurements of light and colour
allow us to some objective comparisons for the study of the temporal dynamics
of the landscape.
|
|
 |
 |
 |
 |
Hue, saturation and luminance parameters can be represented throughout a time interval, along with - on demand - a fourth parameter corresponding to the value of colour clusters dispersion. This last information is not displayed here, it concerns hue - saturation couples, but it reduces to a single value the dispersion that generally can be visually more completely appreciated on the bidimensional hue-saturation diagram. |
 |
luminance
saturation
|
 |
luminance saturation
|
 |
Hue values belong to a mono dimensional topologically circular space,
so one can link those to angles as done in the hue-saturation diagram.
Here the ribbon associated to the hue’s evolution is to be viewed as a
cylinder where upper and lower boundaries meet.
Through some examples, we will try to see how several light "ambiences"
could be translated graphically.
 |
 |
In a 24 hours series, the night-day transition is obviously marked by
luminance variations, however it must be noted that a low luminance sky
can correspond with :
- saturated to very saturated colours (0h,2h, contrasting with 6h),
- variable hues.
Another example
 |
 |
If we consider the ground, look how the presence of the ground with tinted colours (16 h) or poorer (14 h and 6 h) can be shown. On the diagram - at corresponding times - the saturation of the ground goes from a low value to a high one, and to a low value again. The nebulosity is high on the 6 hours image because no cloud intervenes between the sun and the ground which occupies all of the foreground. The latter benefits from direct lighting.
At 6 h, this lighting is too low and filtered to highlight the colours of the ground.
At 14 h, the lighting should be sufficient (see ground luminance diagram),
but the presence of cloud cover influences the richness of the colours
; come to that, it gives a misleading feeling : the ground seems darker
at 14 h than at 16 h.
The ground aspect passes from red to blue dominance : this is due to
the combined effects of an additional snow cover and to lighting.
 
 |
 |
 |
 |
The sudden and significant increase of the ground luminance is typical
of a snow fall. One remarks that the period that follows leads to more
constant and low saturation values for the ground.
|
 |
In good or bad weather, the sky presents various aspects. In cloudy
weather, the layer of clouds can be made of a relatively standard grey
or - as for the third image - weakly tinted. If one considers these 3 images,
one would like to be able to differentiate the second one that shows a
typical phenomenon of meteorological inversion between the fjord and the
ice cap in the background. This light "ambience" is characteristic of bad
weather on the fjord and its banks and sunny weather on the ice cap. In
fact, the first two images are very similar according to hues. If we consider
the sky in the whole, luminances are not radically different.
Here we have to further consider colour repartition in the image : a
grey that shows nothing particular is often something like a luminance
decreasing that - from the top to the bottom of the sky - decreases first
heavily, then softly, and finally eventually dimly increases (image a)
; On the other hand, the meteorological inversion sky corresponds to a
heavy increase of luminance as the sky approaches the horizon (image b).
Here, it is the pattern of the curve that is significant, and not the values
themselves.
We can compare this to what happens with a blue sky, as we then also
have a decrease in luminances from the top to the bottom. One can generally
see that the increase of the luminance is quite uniform in such cases,
contrasting with the precedent case (image d is not chronologically near
the three other).
a 
b 
d 
One can come back to the hue-saturation diagram to simultaneously represent
colour-points corresponding to different areas and moments in time. To
make this readable, one can make clear the two kinds of relations characterising
such a configuration of points.
 |
|
 |
|
 |
If one links – for each considered area – points that temporally follow, one highlights the dynamics of these and one can compare concurrent dynamics. |
 |
 If one links – for every instant – points of areas one highlights a pattern characterising the light “ambience” of each of those single moments |
 |
|
Images 4 and 5 show an evident similitude. Image 5 is on the whole a little more red and saturated than 4 but the dynamics and sky-ground relations are very similar. This corresponds to patterns with the same size and direction, which differ simply by a translation that combines a soft hue shifting and a centrifugal effect toward saturation.
It seems less evident for the pair 1 - 2 : here sky - ground relations are comparable, corresponding to near orientations, but the dynamics of the colours of image 2 - poorer than in image 1 - leads to a kind of reduction.
Image 3 shows on original pattern, due to the grey dominance.
With images 6 and 7 one can observe that "ambiences" can feel similar
even if the sky is blue in one (image 6) and largely covered by clouds
in the other one (image 7). But here the shape of the pattern is mostly
influenced by the ground : on image 6 like in image 7, a large proportion
of the ground surface is dark and dominantly red-tinted. The point of the
diagram that summarises ground colour information accounts for that fact,
but in such a domain human perception does not allow the same weighting
to elements of different luminances. The snow that has melted on image
7 has then, by uncovering a large part of the red ground, increased the
proportion it represents but not the perceptual impact of this colour.
 |
One can search the base
looking for images according to - elements of image - objects of landscape based on their occurrence but also based on their extent in image and on comments that have been added Search : snow, cloudy,
|
The icing of the Loven East Glacier (Spitsbergen, 79° North)
Recurrence of the icing (Loven East Glacier)
They are the result of the action of water circulation underground and
through the snow cover. It is particularly interesting to survey their
evolution in Spring, when the water begins to transform the snow cover
to ice.
The icing of the Loven East Glacier (Spitsbergen, 79° North)
Before this summer aspect arrives, the icing of around 1 km2
is
the object of very important evolutions, at different scales.
Candles of ice (micro-local scale)
When the water infiltrates the snow, it transforms the snow crystal
into candles of ice : the cover becomes blue and the crystals grow, building
some evolved blisters of ice.
Blisters of ice (local scale)
Blisters collapsed
 |
 |
 |
Evolution of the icing during three months :
from May to August. The icing is built and destroy by the water action.
|
If one surveys the total area by a systematic photographic observation,
it is possible to detect daily the presence of water and the importance
of the process by measuring the blue surface on the total area.
Evolution of the icing between May 1st and May 4th 1981
By systematic photo-survey we could understand the general evolution
of the icing, but also the local (blisters) and micro-local (nadles) of
this particular ice.
When the Spring is advanced, the subcover flow becomes aerial : it transforms
and finally partially destroys the icing : the blisters bubbling when the
flow is under the snow grow until they suddenly collapse. Such phenomena
will be better understood through systematic photographic survey.
BORODKINE R., 1999, "Image, apprentissage, appropriation visuelle
de l’espace : approche et implications", actes des 4es rencontres de Théo
Quant, Besançon : 11-12 février 1999, 18 p.
BORODKINE R., 2000, "Perceptual grouping in classificatory constraints",
IFCS'2000, Namur, Belgium, july 11-14.
GRISELIN M., 1991 : “ Les marges glacées du glacier Loven Est,
Spitsberg ”, in Proceedings acts of the 1st International Symposium
Glacier Caves and Cryokarst in Polar Regions, Madrid, Oct. 1990,
pp. 35-67.
GRISELIN M., 1992 : “ Small Polar Glacier Hydrology in Svalbard ”,
in Proceedings of the Arctic research seminar, FFN, Oslo, pp. 89-101.
GRISELIN M., 1992: “ In the Depths of a Small Polar Glacier ” in Proceedings
Acts of the 2nd International Symposium Glacier Caves and Cryokarst in
Polar Regions, Miedzigorge, Pologne, février 1992, pp. 51-63.
GRISELIN M., MARLIN Ch., MOREAU L. et DEVER L., 1995 : “ Hydrology
and Geochemistry of the Loven East Glacier (Svalbard)”, Actes du 3e symposium
Cavités glaciaires et cryokarst en régions polaires et de
hautes montagnes,, article bilingue français-anglais, (pp. 61-76).
GRISELIN M. et MARLIN M., 1998 : “ Origin of the Water Circulation
in the moraine plain of the Loven East Glacier (Spitsbergen) ”, Proceedings
of the 4th International Symposium Glacier Caves and Cryokarst in Polar
Regions, RudolfsHutte-Salzburg, Autriche, 1-6 sept. 1996, Salzburger Geographische
Materialen, Heft 28, pp. 61-71.
GRISELIN M. et ORMAUX S., sous presse : “ Analyse systémique
du paysage visible à partir de photographies au sol : exemple du
bassin Loven Est, baie du Roi, Svalbard ”, actes des 4es rencontres de
Théo Quant, Besançon : 11-12 février 1999, 10 p.
GRISELIN M. et ORMAUX S., sous presse : “ Les nouveaux médias
pour une nouvelle culture du paysage ”. Colloque International Multimédia
et construction des savoirs, Besançon-Belfort-Montbéliard
25-29 mai 1999, 10 p.
GRISELIN M. et ORMAUX S., 1999 : “ Paysage et multimédia ”.
Colloque International Multimédia et construction des savoirs, Besançon-Belfort-Montbéliard
25-29 mai 1999, CD-Rom.
