as a PDF

Transcription

as a PDF
An Automatic Method for Vine Detection in
Airborne Imagery Using Wavelet Transform
and Multiresolution Analysis
Thierry Ranchin, Bernard Naert, Michel Albuisson, Gilbert Boyer, and Par h a n d
Abstract
The problem of automatic vine detection from airborne
Infrared Color (IRC) images is addressed. An automatic method
was designed, based on a separate modeling of the textural
and spectral information extracted from the data. A multiresolution analysis associated with a wavelet transform allows
the separation of spectral and spatial information. After a
description of the vine detection problem, the automatic
method is described. An example application is proposed and
a quantitative evaluation of the results achieved is presented.
The analysis of the results demonstrates the efficiency of the
method. The square system of planting is the main source of
errors. But this growing technique is now nearly abandoned
in France due to mechanization. Thus, this method may help
photointerpreters in the updating or definition of agricultural
registers of vineyards as defined by the European Commission.
Background
The automatic characterization of heterogeneous environments with regular structures are of interest for cultures such
as orchards, vineyards, and market gardens, and is of tremendous importance for the establishment and the survey of
national or European registers. The diversity of objects that
compose these environments (plants, soil, shadow, grass,
stones, and organic residues) makes it a very complex process.
This paper focuses on automatic vine detection. This problem
is representative, at an easily accessible scale in the field, of the
complexity of these environments. Only a few studies have
been conducted in the field of remote sensing for vine detection
(Riom, 1992).
Boissard (1972; 1973) has shown that the textural and
structural elements in avine parcel can be accessed by the Fourier transformation of airborne photographs. These elements
lead to the characterization of the stocks of vine and of their
structure (orientation of the rows, distance between the stocks,
dimension of the stocks, and leaves). Using spectral and temporal parameters with a very high spatial resolution, Debusshe
(1975),Josa (1978),Sentex (1984),and Naert (1985)performed
comprehensive analyses of some elements of the vine states:
physiological states, growing methods, and type of vine. These
T. Ranchin and M. Albuisson are with the Groupe TBlBdBtection & ModBlisation, Centre d'Energbtique, Ecole des Mines
de Paris, BP207, 06904 Sophia Antipolis cedex, France
(thierry.ranchin8cenerg.cma.fr).
B. Naert and G. Boyer are with the Laboratoire INRA, Maison
de la T616dBtection, 500 rue Jean-FranqoisBreton, 34093 Montpellier cedex 5, France.
P a Astrand is with the Space Applications Institute, MARS
sector, Joint Research Center, 21020 Ispra, Italy
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
studies show that the automatic recognition of these pararneters can only be achieved when the detection of vine parcels is
effective. When spectral information alone is considered, the
observed parcels can be confused with non-cultivated soils or
fallow soils. Depending on the cultivation technique of the
vineyard, confusion is possible with orchards or plantings of
asparagus, when only textural information is considered.
Hence, effective detection can only be achieved by approaches
using jointly texture or structure and spectral information. If
temporal information is available, detection will be better. A
vineyard is a patchwork of objects whose signatures change
with the sun position, the season, the orientation of the rows of
vines, the type of vine, and the growing method. For a spectral
and temporal study and for characterization of the vines, the
detection of vines is essential.
As proposed by Boissard (1972; 1973),the use of the Fourier transform was tested by the present authors. But, even if
fast algorithms exist, the use of directional filtering for detection of rows of vines in the Fourier domain of images is very
demanding in computational resources. The Fourier transform
makes some strong assumptions regarding the original image.
The image needs to be periodic, to represent stationary phenomena, and, for computational reasons, to be preferably a
power of 2 in size. But an image is neither periodic nor stationary, and the power of 2 in size limits the application of the Fourier transform. Nevertheless, the Fourier transform was
successfully applied to images. The following procedure was
applied to perform a directional filtering of the image for line
detection:
a Fourier transform is applied to the original image (Figure 1);
the filter is designed manually (Figure 2) on the amplitude
spectrum, by selecting the frequencies to preserve (Figure 3);
this first filter is converted into an optimal filter which is applied
to the amplitude spectrum of the image; and
the inverse Fourier transform is applied to the filtered spectrum
and the filtered image is reconstructed (Figure 4).
This method is efficient but, in addition to the previous
restrictions, it should be stressed that the design of the filter is
done by the user, which strongly influences the results. This
point prevents the automation of the method. Due to these different remarks, this method was not further employed and an
alternative one was designed.
More flexible mathematical tools were chosen: the wavelet
transform (WT) and multiresolution analysis (MRA). These
Photogrammetric Engineering & Remote Sensing
Vol. 67, No. 1, January 2001, pp. 91-98.
0099-1112/01/6701-91$3.00/0
O 2001 American Society for Photogrammetry
and Remote Sensing
lanuary 2001
91
I
Figure 1. Original image. The pixel size is 25 cm.
Figure 2. Amplituae spectrum of tne touner transform computed from Figure 1.
tools allow a hierarchical description of the information. The
textural information is accessed through the use of the WT and
MRA, and the spectral information is accessed through a classification scheme.
In the field of the analysis and processing of remotely
sensed images, a few works were done using these tools and concerning the analysis of the characteristic scales in geology (Besnus et a]., 1993),in the urban domain (Ranchin and Wald,
1993a), or in oceanography (Ranchin and Wald, 1993b). The
Figure 3. Mask defined mar .--..,
directional filter.
... order t~ --...,
~ t the
e
Figure 4. Reconstructed image computed from the applica
tion of the directional filter on Figure 1.
reduction of speckle in Synthetic Aperture Radar (SAR) imagery was proposed using degradation of the spatial resolution of
the image (Proenca et al., 1992),by selection and filtering of the
most speckled scales (Ranchin, 1993; Ranchin and Cauneau,
1994),by application of the shrinkage method on the wavelet
coefficient images (Eldridge and Lasserre, 1995; Horgan, 1998),
or through the use of a recursive scheme (Dong et al., 1998).
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
The restoration of blurred images was studied in the case of airborne images (Bruneau et al., 1991).The benefits of using these
tools for data compression were demonstrated from the point of
view ofperformance (Antonini et al., 1992; Singh, 1999), and
the European Space Agency has proposed the distribution of a
CD-ROM of Meteosat images using such a kind of compression.
SAR data compression was also explored with significant good
results (Moureaux et al. 1995; Parkes and Clifton, 1999).An
automatic method for registration of images was proposed
(Djamdji et al., 1993). The detection and selection of ground
control points (GCP) is done using the WT,and the computation
of the transformation model is achieved through successive
approximations in a coarse to fine approach. This method was
deeply improved by Blanc and Wald (1998) by taking into
account the local deformation of images.
These tools were used for the analysis of satellite-derived
irradiance maps and for the characterization of the spatial and
temporal structures in the irradiance fields (Beyer et al., 1995).
A characterization of the local roughness of digital elevation
models by a wavelet-based estimator (Datcu et al., 1996)and
surface determination for digital terrain reconstruction and
ortho image formation (Tsay, 1998)were improved by the use of
the WT. The scale analysis of surfaceproperties of sea ice (Lindsay et al., 1996)or of sea ice drift (Liu and Cavalieri, 1998)were
also investigated. These tools were also applied to the fusion of
images and raster-maps of different spatial resolutions by
encrustation in the case of the construction of the European
solar radiation atlas (Wald and Ranchin, 1995),to the merging
of images for updating urban maps (Ranchin and Wald, 1996),
and to the merging of SPOT and SAR images (Mangolini et al.,
1993).They are the basis of the ARSIS method (from its French
name "am6lioration de la rbolution spatiale par injection de
structures") which is the subject of a patent (Mangolini et al.,
1992).This method allows the improvement of the spatial resolution of images up to the best available in a set of images with
different spatial and spectral resolutions, and the preservation
of the spectral content of original images. In the case of SPOT
imagery, this method was shown to give the best available
results in terms of preservation of the spectral content of original images (Wald et al. 1997).Thus, the synthesized images can
be used for applications where high spatial and spectral resolutions are necessary. The complete description of the implementation for this case is provided in Ranchin and Wald
(2000).
Couloigner andRanchin (1999; 2000) demonstrate the benefits of these tools for the semi-automatic extraction of urban
networks. Some potentialities of the WT and MRA were also proposed by Ranchin and Wald (1993a). Ranchin (1997; 1999)
explored the benefits of wavelets for environmental modeling.
MRA and the WT were also used for the detection of settlements
in a water resources management context (Staudenrausch et al.,
1999)and for the automatic interpretation of buildings from
aerial images (Shao, 1993).Interesting studies on the multiscale
properties of images (Hu eta]., 1998)and on texture analysis and
classification using wavelets (Zhu and Yang, 1998)were also
achieved. This list is far from being exhaustive because the
recent use of these tools in remote sensing and their potentials
provide new applications every day.
An Automatice Method for Vine Detection
The Wavelet Transform and Multlresolutlon Analysis
The association of the wT and MRA leads to a powerful and comprehensive analysis and processing of remotely sensed images.
The concept of MRA introduced by Mallat (1989) derives from
the Laplacian pyramids (Burt and Adelson, 1983).
Figure 5 is a very convenient description of MRA and more
generally of pyramidal algorithms.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
of the original
Difference of informatton
between two successive
appmx~mationsmodelised
by the wavelet coefficients
Original image
Figure 5. Pyramid representing the multiresolution analysis combined with the wavelet
transform.
Figure 6. Strategy for vine detection.
The base of the pyramid is the original image. Each level of
the pyramid is an approximation of the original image computed from the original one. When climbing the pyramid, the
successive approximations have coarser and coarser spatial resolutions. The theoretical limit of an MRA is one pixel representing the mean of the original image. Due to some physical
constraints, this limit is never reached. The base of the pyramid
can also be considered as an approximation of the landscape
measured by the sensor.
Associated with MRA, the WT allows the description of the
differences existing between two successive approximations
of the same image (i.e., of two successive levels of the pyramid,
by its wavelet coefficients).If the process of MRA is inverted,
the original image can be exactly reconstructed, from one
approximation and from the different wavelet coefficients.
These tools allow a hierarchical description and modeling of
the information contained in the image. The present method
uses the "B trous" algorithm (Dutilleux, 1987). For each iteration, one context image and one wavelet coefficient image are
computed. They have the same number of pixels as the original
one, and can be compared pixel by pixel to the original one.
The wavelet coefficient image represents the characteristic
scales of the original image with no privileged direction.
The Propmd Strategy
Figure 6 describes the strategy for the automatic detection of
vines. The strategy is as follows:
were achieved by fitting the histogram by a generalized
Gaussian law. The values close to zero (central peak of the histogram) represent noise or a very small variation of the image.
The high values of the histogram (left and right parts of the histogram) represent the strong variations of the image, i.e., wellmarked structures as borders between different elements.
The method of line detection is based on this interpretation
of the histogram. Hence, a thresholding of the wavelet coefficients image of the relevant structures will enhance the wellmarked structures and show the lines of vines. The threshold
was determined empirically to be 0.7 uwhere uis the root mean
square of the wavelet coefficients image between one and two
meters of the image. This threshold depends on the application
and on the size of the structures to be detected; it shouldbe fitted according to the objectives.
The algorithm of line detection is item (2) of the strategy
scheme (Figure 6) and may be described as follows:
I
L
Y
Y
Y
E
J
I
"
I
m
m
Figure 7. Typlcal histogram of the wavelet coeffic~entsImage representing the information at a
given scale (here between one and two meters)
for a natural image.
(1)The original infrared color (IRC) image is decomposed into
conven6onal red, green, and blue layers. ~ c c o r d i nto~the
usual processing of airborne photographs (Naert, 1977), the
vegetation is mostly enhanced in the green layer which contains the infrared information (Boureau and Naert, 1985).
using the WT is applied to this layer.
(2) Line detection is applied to the wavelet coefficients image
representing the structures to be detected. The choice of this
image depends on an a priori knowledge of the distances
between the rows of vines, and can easily be tuned.
(3) A textural index is derived from the results of this line
detection.
[4) Approximations of the original data computed through MRA
and the textural index are used to derive a supervised classification of the whole image.
(5) Results of the classification are analyzed in a G I environment
~
in order to provide the results on a parcel basis. Then a combination of the parcel shapes and the results of the classification
is performed. A threshold is applied to determine if the parcel
is a vine parcel. All the parcels with a value in surface detected
greater than 75 percent of the reference surface are considered
as vine. This threshold was determined empirically from the
individual parcel results.
=
The richness of the details of high spatial resolution imagery enables the detection of individual stocks but prevents the
effective detection of the parcel of vines. On the one hand, the
high spatial resolution imagery is necessary for accessing the
texture of vine areas. On the other hand, such images are too
complex for accessing the spectral signature of a vine parcel.
This makes difficult the definition of an automatic detection.
To avoid this problem, using the information at a selected scale
focuses on the structures of the relevant dimensions; the
emphasis is put on the lines of vines, and the spectral signature
of vine parcels is well represented by smooth approximations
of the original image. Additionally, to obtain this textural information, it is necessary to start from an image with a pixel size at
least two times smaller than the size of the structures to detect
(Shannon theorem; see, e.g., Papoulis (1987)).
Line Detection and Textural index
The wavelet coefficients images have histograms that are centered at zero and present avery sharp distribution. Figure 7 is a
typical example of the histogram of the wavelet coefficients
image for a natural image. This histogram, which is a representation ofthe probability density function (pdf)of the image, can
be adjusted to a theoretical pdf obeying a Laplacian or a
Gaussian law. Following Antonini et al. (1992),the best results
an MRA using the WT is applied to a single channel of the image
(usually the channel best representing the vegetation),
the root-mean-square uis computed for the wavelet coefficients
image comprised between one and two meters,
all the wavelet coefficients with value lower than 0.7 u are set
to 0, and
all the wavelet coefficients with value greater than 0.7 r a r e set
to 255.
A texture index is derived from the line detection image by
computing the number of pixels with a value equal to 255 in a
moving window of size 9 by 9. This index will be used as one
input layer in the subsequent supervised classification.
Example of Application
The area of interest is situated on the district of Villeneuve les
Maguelone (HBrault, France]. It comprises the Viticultural
Experiment Station of INRA, called "Domaine du Chbpitre," surrounded by representative and exhaustive samples of the languedocian vineyard. In 1984, this station exhibited structures
of plantations that are still (i.e., rows 2 m apart with a distance
of 1.2 m between stocks, 2 m by 1.3m, and 2.5 mby 1.2 m) in
bush vines (goblet-prunedvines) or lateral cordon-constrained
and trellised vines with numerous different varieties of vines.
Around the experiment station, the parcels are not homogeneous. Many of them are planted in bush-vines with a square
system of planting (1.5 m by 1.5 m or less); a path exists every
three or four rows, for the treatment tractors. The image was
acquired using an IRC (inhared color) emulsion, at a scale of
1:5000 on 16 October 1984 and is composed of urban area (more
or less diffuse), vineyard (the "Domaine du Chbpitre" represents 20 hectares of the vineyard), fallow land, asparagus, and
orchards.
Plate 1presents an example of application of the complete
strategy. The inter-row distance of vines is between one and
two meters. Then, the line detection algorithm has been
applied to the wavelet coefficients image describing the information between one and two meters. The spatial resolution of
the approximations (computed by application of MRA) used for
the spectral description of the area is 8 meters in this case. The
masking was applied to the results of the classification. The
surfaces of the resulting parcels are then compared to the reference surfaces. If the individual difference for each parcel is
lower than 25 percent in area, the parcel is classified as vine. If
the difference is greater than 25 percent, the parcel is classified
as non-vine. In the next section, the results for the whole image
are summarized using error matrices and indexes.
Results and Discussion
The evaluation of the results was first achieved for each parcel.
In order to obtain an indicator for the whole process, the
approach proposed by Congalton (1991) was used. An error
matrix was computed for the image. It contains two classesPHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
TABLE2. ACCURACYINDICES
DERIVED
FROM TABLEI
User's accuracy indexes
Producer's accuracy indexes
Overall accuracy index
KAPPA parameter
Plate 1 . Example of the strategy for automatic vine detection. The upper left corner is an extract of the reference
image. The pixel size is 0.25 m and the image was acquired
using an infrared color emulsion at the scale of 1:5000.
The upper right corner is the result of the line detection
algorithm using the wavelet transform. The lower left corner
is the superimposition of the original image and the parcel
information. The lower right corner is the combination of
the results of the classification scheme and the parcels
information.
vine and n o n - v i n e where columns describe the reference
data and rows indicate the assigned land-cover categories. The
followingindices were calculated:
the user's accuracy indices (an inverse measure of the commission errors),
the producer's accuracy indexes (an inverse measure of the
omission errors],
the overall accuracy index, and
the KAPPA parameter (directly computed without any use of the
KHAT estimator].
Table 1 presents the error matrix computed for the whole
image and Table 2 presents the calculated statistical indices.
The overall accuracy has to be considered with the "producer's accuracy" and the "user's accuracy." The overall accuracy for the whole image is close to 82 percent. It excludes the
omission and commission errors. The producer's accuracy
indicates that approximately 74 percent (26 percent omission
of vine) of the vines are well-classified and that approximately
95 percent of the non-vines are well classified. The user's accuracy indicates that approximately 96 percent (4 percent commission or mix-up of vine) of the vine parcels are actually vine
and that approximately 68 percent of the non-vine parcels are
TABLE1. ERRORMATRIX
FOR THE
IMAGE WITH A
Vine
Non vine
0.96
0.68
0.74
0.95
0.82
0.64
actually non-vine parcels. The KAPPA parameter is a measure of
how well the classification agrees with the reference data, and
takes into account the omission and commission errors. This
parameter is usually considered as the one reflecting the global
quality of the classification. For this image, it reaches approximately 64 percent. This is a fairly good result compared to the
usual results obtained by photointerpretation.
In order to analyze the contribution of the different growing methods, extended error matrices were computed and the
producer's accuracy indexes derived first for different types of
planting (Table 3) and vine-growingmethods (Table 4 and 5).
This area comprises two main types of planting existing in this
area: the tying up types composed of stocks growing on wires,
for which the leaves of the different stocks constitute a continuous line, and the bush or goblet pruning which corresponds to
stocks planted individually without any physical link as wire
in the tying up mode. Depending on their inter-rows distance,
the leaves can give an impression of continuity in lines. There
are two modes of planting for goblet pruning. Figure 8 displays
the square system of planting and the usual goblet pruning. The
differences are in the inter-line and inter-row distances.
Table 3 shows that the main sources of error are due to the
young and badly maintained vines and to the square system of
planting. The young vines are very difficult to detect even using
photointerpretation. The badly maintained vines are presenting missing stocks, holes in rows, and do not reach the 75
percent of vine surface occupation. The vines planted in rows
are very well detected (less than 4 percent of omission). Table 4
shows that the method allows a good detection, whatever the
growing method, when the vine is planted in rows. The square
system of planting is the main source of error in our strategy for
automatic vine detection. These parcels represent 43 percent of
the omission.
Table 5 shows the results for the five different classes of
squares that are present in the image. The regular square system of planting corresponds to stocks planted on a regular grid
in lines and rows spaced at 1.5 m that gives an impression of
squares. No link exists between the stocks. The other classes are
composed of the same regular grid but with one row out of two,
three, four, or six missing. The missing lines in the parcels correspond to the pulling up of stocks, allowing the use of tractors
for exploitation of the vines. Table 5 shows that the square system of planting itself presents an omission of 23 percent. The
interpretation of the square with one row out of three missing is
difficult, because the number and size of parcels is not statistically significant. The square with one row out of two missing is
PIXELSIZEOF 25
CM AND
PROCESSEDUSINGTHE PROPOSED
STRATEGY
Ground Truth
Detected areas
Vine
Non vine
Total
Vine in square meters in number of parcels
Non vine in square meters in numbers of parcels
Total in square meters in number of parcels
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
January 2001
95
ACCURACYINDEX BASED
UWN THE TYPE
OF PLANTATION
INFORMATION FOR ME IMAGE PROCESSED
WITH ME
TABLE3. EXTENDEDERRORMATRIXAND PRODUCER'S
PROPOSED
STRATEGY
--
Vine in m2 in parcels
Non vine in mZin parcels
Total in mZin parcels
Producer's accuracv index
-
Rows
Square
197131
33
7489
2
204620
35
0.96
79164
22
60714
15
139877
37
0.57
Young or badly
maintained
No
information
Total
EXTENDED ERRORMATRIXAND PRODUCER'S
ACCURACYINDEX BASED
UPON WE DETAILED
TYPE
OF VINE-GROWING
METHODFOR THE PLANTATION
IN ROWS
TABLE4.
---
--
5 i n g UP
Vine in m2 in parcels
Non vine in mZin parcels
Total in mZin parcels
Bush or goblet
pruning
157219
23
7489
2
164708
25
39912
10
0.95
1
a
0
0
39912
10
Producer's accuracy index
(a)
(b)
Figure 8. (a) Usual bush or goblet pruning. (b) Square system of planting.
very close to a classical row system of planting. But the distance between two successive rows is 3 m. The algorithm was
tuned to detect vines spaced of between one and two meters.
Hence, the omission of more than 54 percent, is due to this tuning. One way of detecting this type of growing method parcels
is to process the wavelet coefficients image representing the
corresponding characteristic scales. The omission decreases
from 54 percent to 49 percent when the number of missing rows
decreases. With the different spacing due to missing rows, the
algorithm is able to detect the rows, but the masking leads to a
surface occupation inferior to 75 percent. The part of the parcels covered by vines is well detected but no connection is
made between these sub-parcels. One way of resolving this
problem can be an adaptation of the algorithm by the computation of another textural index focusing on the characteristic
scales of this process.
Conclusions and Perspectives
In this paper, an automatic method to detect the vinelnon-vine
areas on aerial infrared photographs was described. The developed method allows detecting the vine areas in the test district
with a Kappa coefficient of 0.64 in the case of a 1:5000-scale
airborne IRC image.
The pixel size of 50 cm seems to be a limit to properly
detect the vines planted in squares. However, this technique of
Square,
Vine in mZin parcels
Non vine in mZin parcels
Total in mZin parcels
Square
1 row12 missing
37093
6
11112
4
48204
12726
15133
10
11
6
5
27859
culture has been abandoned due to mechanization. The proposed method is able to detect the two or three rows of planted
vines but has difficulty to link them through the grass lines.
Other tests have also been achieved with 1:20,000-scale
images and a pixel size of 50 cm and over different vineyards
in Spain (panchromatic 1:40,000-scale image, pixel size 50 cm),
Italy (panchromatic 1:40,000-scale image, pixel size 1m), and
France in the area of Bordeaux ( L R 1:25,000-scale
~
image, pixel
size 60 cm). The results obtained, without any modification of
the method, on these images show a lower accuracy. This
emphasizes the need for an adaptation of the method to fit the
imagery, the type of vine growing method, and the inter-row
distance present in the area under study.
This method is fully automatic. Its applicability to the
detection of vines parcels was demonstrated for a 1:5000-scale
IRC image. The level of detection of the vinelnon-vine parcels is
fairly good (Kappa parameter equal to 64 percent). The choice
of the wavelet coefficients image to process permits an easy
adaptation of the method to the available image and to the
characteristics of the vineyard.
The present results are less satisfactory than those usually
obtained by photointerpretation. This method is a first step to
Square,
1 row13 missing
Square,
1 row14 missing
1518
1
6758
1
8276
2
20083
6
20391
4
40474
10
Square,
1 row16 missing
Producer's accuracy index
0.77
96
January 2001
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
an efficient automatic method for vine detection. It constitutes
a good framework and will be improved by further investigation of the different ways of representing the textural information of vines.
This method can presently be used as a support for photointerpretation, allowing a first step in the processing of vine
detection. The photointerpreters can then focus on the parcels
that are not detected as vines. The detection of changes in the
use of parcels through time can also be conducted through comparison ofresults computed on different dates.
-, 2000. Mapping of urban areas: A multiresolution modeling
approach for semi-automatic extraction of streets. To appear in
Photogrammetric Engineering & Remote Sensing, 66(7):867-874.
Datcu, M., D. Luca, and K. Seidel, 1996. Wavelet-based digital elevation
model analysis, Proceedings of the 16th Symposium of EARSeL,
Integrated Applications for Risk Assessment and Disaster Prevention for the Mediterranean, 20-23 May, Ta'Xbiex, Malta, pp.
283-290.
Debusshe, G., 1975. Reconnaissance des c6pages par les photographies
a6riennes dans les vignobles du haut et bas Minervois, France,
Photo-Interprttation, 75-3.
Djamdji,
J.P., A. Bijaoui, and R. Manihre, 1993. Geometrical registration
Acknowledgments
of images: The multiresolution approach, Photogrammetric EngiThis study was partly supported by the Joint Research Center of
neering & Remote Sensing, 59(5):645-653.
Ispra, through the VINIDENTproject. The authors thank ONMN
Dong, Y., B.C. Fonter, A.K. Milne, and G.A. Morgan, 1998. Speckle
and the Regional Statistics Department of the Languedoc-Roussuppression using recursive wavelet transforms, International
sillon (S.R.S.A.) for the statistical description of the France
Journal of Remote Sensing, 19(2):317-330.
vineyards, the INRA team of the "domaine du Ch3pitre" for the
Dutilleux, P., 1987. An implementation of the "algorithme & trous" to
description of the growing techniques, and the GUTLAR for furcompute the wavelet transform, Proceedings du congrds ondeletnishing the airborne photographs. The authors want to thank
tes et m&hodes temps-Fquence et espace des phoses, Marseille,
the anonymous reviewers for their fruitful comments.
14-18 September, Marseille, France, Springer-Verlag Editors,
pp. 298-304.
Eldridge, N.R., and M. Lasserre, 1995. Wavelet transform analysis and
References
speckle filteriog of synthetic aperture radar (SARI sea ice imagery,
Antonini, M., M. Barlaud, P. Mathieu, and I. Daubechies, 1992. Image
Proceedings of the Seventeenth Canadian Symposium on Remote
coding using wavelet transform. BEE knsactions on Image PmSensing, 13-15 June, Saskatoon, Saskatchewan, Canada, pp.
cessing, 1(2):205-220.
161-167.
Besnus, Y., J.C. Pion, M. Raffy,G. Ramstein, and H. Y6sou, 1993.b i t e - Horgan, G., 1998. Wavelets for SAR image smoothing,Photogrammetric
ments d'images appliqu6s & la cartographie des formation dun&
Engineering & Remote Sensing, 64(121:1171-1177.
de l'erg Akchar (Mauritanie),apport des donn6es Landsat TheHu, Z., Y. Chen, and S. Islam, 1998. Multiscaling properties of soil
matic Mapper et de l'analyse par ondelettes d'une image SPOT
moisture images and decomposition of large- and smd-scales
panchromatique, Actes du colloque de Toulouse. Outils microfeatures using wavelet transform, International Journal of Remote
informatiques et t6166tection de 1'6volution des m7ieux: bisensing, 19(12):2451-2468,
sidmesjourn6es scientifiquesdu r6seau de t616d6tection de I'URFF,
13-16 November 1990, Toulouse, France, Presses de lYUniversit6 Josa, R., 1978. T616d6tection et p6dologie. Cornparaison d'kchelles de
prises de vues sur la mission photogmphique IRC du Montpellibrdu Qu6bec, Sainte-Foy (Qubbec),pp. 253-269.
ais 1976. Rapport de stage GDTA-SES W 339, Groupement de
Beyer, H.G., C. Costanzo, and C. Reise, 1995. Multiresolution analysis
DBveloppement de la Teledetection A6mspatiale et hstitut
of satellite-derived irradiance maps: An evaluation of a new tool
National de la Recherche Agronomique bditeurs, Toulouse, France,
for the spatial characterization of hourly irradiance fields. Solar
133 p.
Energy, 55(1):9-20.
Lindsay, R.W., D.B. Percival, and D.A. Rothrock, 1996. The discrete
Blanc Ph., and L. Wald, 1998. Validation protocol applied to an autowavelet transform ans the scale analysis of the surface properties
matic co-registration method based on multi-resolution analysis
of sea ice, BEE Tkansactions on Geoscience and Remote Sensand local deformation models. International Archives of Photoging,
34(3):771-787.
mmmetry and Remote Sensing, 32(Part 2):ll-19.
Liu, A.K., and D.J. Cavalieri, 1998. On sea ice drift from the wavelet
Boissard, P., 1972. Etude du spectre des frbquences spatiales d'une
analysis of the Defense Meteorological Satellite Program (DMSP)
photographie.
Photo-Interpfitation, 5(4):21-29.
Special Sensor Microwave Imager (SSMII)data, International Jour, 1973. Recherche d'une aide automatique h 11interpr6tationdes
nal of Remote Sensing, 19(7):1415-1422.
images a6riennes. ADDlication h I'Mude de la v6e6tation. These
Mallat, S.G., 1989. A theory for multiresolution signal decomposition:
de 6octorat de sp6cibt6, Facult6 des Sciences, ~ r l b n sFrance,
,
the wavelet representation, IEEE Mnsactions on Pattern Analysis
100 p.
and Machine Intelligence, 11(7):674-693.
Boureau, J.C., and B. Naert, 1985. Caracthrisation des formations
Mangolini,
M., T. Ranchin, and L. Wald, 1992. Procbd6 et dispositif
naturelles hbt6rogbnes par dbgradation du pixel des satellites 1
pour l'am6Jiomtion de la r6solution spatiale d'images &pmtird'auhaut pouvoir de r6solution, Application & la cartographie des boitres images de meilleure r6solution spatiale, Patent no 92-13961,
sements 18ches, T616d6tection et Gestion des Ressources: l'aspect
20 November 1992.
op6mtionne1, Vol V (M. Bernier, G. Lessard, and P. Gagnon, edi, 1993. Fusion d'images SPOT multispectrales (XS) et panchrotors), Association Qu6b6coise de T616d6tection, Chicoutimi, Quematique (P),et d'images radar. De l'optique au radar, les applicabec, Canada, pp. 311-325.
tions de SPOT et ERS, Cepaduhs-Editions, 111 rue Vauquelin,
Bruneau, J.M., L. Blanc-FBraud, and M. Barlaud, 1991. Op6rateurs de
Toulouse, France, pp. 199-209.
r6gularisation en restauration d'image: calculs et comparaisons,
Moureaux, J.M., P. Gauthier, and M. Barlaud, 1995. Raw SAR data
Actes du treizidme colloque du GRETSI, 16-20 September, Juan
compression using vector quantization, International Journal of
Les Pins, France, pp. 781-784.
Remote Sensing, 16:3179-3187.
Burt, P.J. and E.H. Adelson, 1983. The Laplacian pyramid as a compact
Naert, B., 1977. Analyse et combinaison chromatique appliqu6e & une
image code. BEE lkansactions on Communications,
photographie ballon infra-rouge couleur. Attenuation de l'effet
31(4):532-540.
versant, Photo-Interprtjtation, 3(3):21-29.
Congalton,R.G., 1991. A review of assessing the accuracy of classifica,1985. Caract6risation de la signature spectrale, spatiale et temtions of remotely sensed data. Remote Sensing of the Environporelle des vignes et des vergers. Application &lat616d6tection du
ment, 37:35-46.
sol et de la culture, Actes du Ssmes6minaire international des
Couloigner, I., and T. Ranchin, 1999. Semi-automatic extraction of
signatures spectrales, 16-20 December, les Arcs, France, pp.
urban road network: assessment of the quality, Proceedings of
467-474.
18th Symposium of EARSeL, Operational Remote Sensingfor Sustainable Development, (G.J.A. Nieuwenhuis, R.A. Vaughan, and
Papoulis, A., 1987. Signal Analysis, McGraw-=ll International EdiM. Molenaar, editors), 11May 1998, Enschede, The Netherlands,
tions, Electrical & Electronical Engineering Series, Singapore,
A.A. Balkema, Rotterdam, Brookfield, pp. 309-314.
Malaysia, ISBN 0-07-048460-0, 431 p.
-
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
January 2001
97
Parkes, S.M. and H.L. Clifton, 1999. The compression of raw SAR and
SAR image data, International Journal of Remote Sensing,
20(18):3563-3581.
Proenca, M.C., J.P. Rudant, and G. Flouzat, 1992. Using wavelets to
get SAR images "free" of speckle, Proceedings of the 12th Symposium IGARSS'92, International Space Year: Space Remote Sensing, 26-29 May, Houston, Texas, pp. 887-889.
Ranchin, T., 1993. Applications de la transformbe en ondelettes et de
I'analyse multirbsolution au traitement des images de tblbdbtection. These de Doctorat en Sciences de l'Ingbnieur, Universit6 de
Nice-Sophia Antipolis, France, 146 p.
, 1997. Wavelets, remote sensing and environmental modelling,
Proceedings of the 15th IMACS World Congress on Scientific Computation, Modelling and Applied Mathematics, 24- 29 August,
Berlin, Germany, Volume 6, Application in Modelling and Simulation, pp. 27-34.
, 1999. Wavelets for modeling in remote sensing, Proceedings
of SPIE's 1999 International Symposium on Optics, Imaging and
Instrumentation. Mathematical Imaging: Wavelet Applications in
Signal and Image Processing VIZ, A.F. Laine, M.A. Unser and A.
Aldroubi, editors), 18-23 July, Denver, Colorado, 3813:208-218.
Ranchin, T., and F. Cauneau, 1994. Speckle reduction in Synthetic
Aperture Radar imagery using wavelets, Proceedings of SPIE's
1993International Symposium on Optics, Imaging and Instrumentation. Mathematical Imaging: Wavelet Applications in Signal
and Image Processing, 11-16 July 1993, San Diego, California,
2034:432-441.
Ranchin, T., and L. Wald, 1993a. The wavelet transform for the analysis
of remotely sensed images. International Journal of Remote Sensing, 14(3):615-619.
-,
1993b. Applications of wavelet transform in remote sensing
processing, Proceedings of 12th Symposium of EARSeL, Remote
Sensing for Monitoring the Changing Environment of Europe,
(Peter Winkler, editor), 08-11 September 1992, Eger, Hungary, A.
A. Balkema, Rotterdam, Brookfield, pp. 261-266.
, 1996. Benefits of fusion of high spatial and spectral resolutions
images for urban mapping, Proceedings of the 26th International
Symposium on Remote Sensing of Environment and the 18th
I
Annual Symposium of the Canadian Remote Sensing Society,
25-29 March, Vancouver, British Columbia,Canada, pp. 262-265.
, 2000. Fusion of high spatial and spectra1 resolution images:
The ARSIS concept and its implementation, Photogrammetric
Engineering b Remote Sensing, 66(1):49-61.
Riom, M., 1992. Suivi de la vigne par t616d6tection, Agrombt~orologie
de la vime en France, CCE EUR 13911, Commission des Communaut6s europ6enes, Direction g6n6rale telecommunications,
Industries de l'information et Innovation, Luxembourg, ISBN 92826-3536-8, pp. 143-183.
Sentex, M., 1984. Tblbd6tection: application d la viticulture, D.A.A.
viticulture, Ecole Nationale Supkrieure dlAgronomie,Montpellier, France.
Shao, J., 1993. Wavelet Theory and Automatic Interpretation of Building from Aerial Images, Ph. D. Dissertation, Wuhan Technical
University of Surveying and Mapping, Wuhan, China, 162 p.
Singh, V.K., 1999. Discrete wavelet transform based image compression, International Journal of Remote Sensing, 20(17):3399-3406.
Staudemausch, H., W.-A. Fliigel, T. Ranchin, I. Herlin, G. Rodolfi, M.
Clark, R. Schulze, N. King, D. Tevera, and J. Matondo, 1999. The
development of an innovative computer based "Integrated Water
Resources Management System" for semi-arid catchments: Concepts and first results, Zentralblatt fiir Geologie und Paliiontologie.
Teil I: Allgemeine, angewandte, regionale und historische Geologie, Heft 3-4, pp. 189-212.
Tsay, J.-R., 1998. A new algorithm for surface determination based on
wavelets and its practical application, Photogrammetric Engineering b Remote Sensing, 64(12):1179-1188.
Wald, L., and T. Ranchin, 1995. Fusion of images and raster-maps of
different spatial resolutions by encrustation: Smodhing the
edges, Computers, Environment and Urban Systems, 19(2]:77-87.
Wald, L., T. Ranchin, and M. Mangolini, 1997. Fusion of satellite images
of different spatial resolutions: assessing the quality of resulting
images, Photogrammetric Engineering 6. Remote Sensing,
63(6):691-699.
Zhu, C., and X. Yang, 1998. Study of remote sensing image texture
analysis and classification using wavelet, International Journal
of Remote Sensing, 19(16):3197-3202.
CALL FOR PAPERS
PE&RS Special Issue on
Geospatial Information Technology in KOREA
Remo
sing
n December 2001, the American Society for Photogrammetry a n d Remote Sensing will
ievote its issue of Photogrammetric Engineering a n d Remote Sensing ( P E e R S ) to
2eospatial Information Technology in KOREA.
4uthors are encouraged to submit manuscripts o n "cutting edge" research underway i n
(OREA i n the following areas
Photogrammetry Remote Sensing LIDAR
Radar GIS GPS Biological & geological applications
All aspects of geospatial information technologies
411 manuscripts must be prepared according to the "Instructions to Authors" published
n each issue of PEbRS and o n the ASPRS web site at www.asprs.org. Papers will be
3eer-reviewed i n accordance with established ASPRS policy. Manuscripts must be re:eived by March 31, 2001 to be considered for publication.
Please send completed manuscripts or direct inquires to :
Dr. Woosug Cho, Guest Editor
Department of Civil Engineering, Inha University
253 Yonghyun-Dong Nam-Gu, Inchon, 402-751, KOREA
FAX : +82-32-873-7560, [email protected]
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING