FUSION OF MULTIRESOLUTION AND MULTISENSOR IMAGERY
IN C¸T H¶I, H¶I PHßNG
PHAN TRäNG TRÞNH1, HOÀNG QUANG VINH1,
ANDRÐ OZER2, MAI THANH T©N3
1Department of Geodynamics, Institute of Geological Sciences, NCST, Hoàng Quèc ViÖt,
CÇu GiÊy, Hà Néi.E-mail:
Abstract:
The use of collateral remote sensing datum sets provide with additional benefits of redundancy and complementarily that are required by data integration and data fusion technique. SPOT data have been widely integrated among multispectral sensors especially PAN-XS fusion and PAN-TM fusion using technique such as Intensity-Hue-Saturation (IHS), principal components analysis (PCA) and high-pass filter (HPF). High resolution satellite images is very important for environmental study and detail mapping of the urban areas and coastal zones. LANDSAT and SPOT XS images have advantage of spectral and large dimension. However, their spatial resolution is not sufficient in comparison with IKONOS images that have disadvantage of extreme expensive and limited surface. It is first time, we successfully make in this study the fusion of SPOT MS and aerial photo. The spatial resolution increases 10-20 times in comparing with SPOT MS. Some fusion techniques are used such as Intensity - Hue - Saturation transform (IHS), high-pass filter (HPF). The fusion technique can also integrate among multispectral sensors with different resolution. SPOT data are integrated with RADAR data to enhance feature discrimination. Applying in Cát Hải coastal zone, the quality of SPOT MS imagery are intensively ameliorated, so one can map small road of the villages, the dikes, small salt marsh and shrimp basin. In the other part, thank to the conservation of multispectral resolution, one can more understand the sedimentary transport in shallow and mangrove mapping. In Hải Phòng area, one can map port bridge, container port and evaluate recent industrial affect. SAR images of ERS-1, ERS-2 and RADARSAT satellite analyses allow to extract the wave pattern. Such information has been proved to be necessary to explain the beach erosion areas. Analysis of aerial photos of 1952 and 1993 completed by SPOT-HRV together with LANDSAT-TM satellite images allow to map the erosion and accumulation areas and to highlight the relation between these coastal geomorphologic processes and the wave patterns.I. INTRODUCTION
The advantage of satellite data is well known. It provides at any scale, up to date, inexpensive information on digital form. Many tools, which allow to extract environmental data from satellite imagery, have been developed. With the rapid development of remote sensing technologies, such as the development of the new generation imaging sensors, leading to enhanced performance at a cheaper price, multisensor systems have become a reality in a growing quantity of applications. Earth imaging, and medical imaging are some of the areas benefiting from such systems in addition to the battlefield applications for which they were first developed. Larger and spectrally more independent sensor arrays provide for increased spatial resolution and better spectral discrimination of the image data available for these applications. However, implementation of such sensor arrays has resulted in a significant increase in the raw amount of image data which needs to be processed. Most of the conventional image processing software has been designed for optimal operation on single images and their application in multisensor arrays must be backed up by a large increase in computational power. An alternative of this costly solution is in the form of image fusion algorithms which provide with an effective way of reducing the total amount of information presented without perceptual loss of image quality or content information. Other advantages of image fusion such as improving situational awareness and night pilotage assistance have also been documented in literature. Although they achieve data amount reduction, image fusion algorithms still operate on very large input information sets and as a result their computational complexity can be prohibitively high for fast, real or near-real time vision system operation. It is, therefore, imperative to develop simple and efficient fusion techniques if the implementation of image fusion is to become a reality. Between them, image fusion systems can be differentiated according to the processing level at which information fusion takes place. Generally, fusion can be at symbol, object and pixel level. During the past decade, a number of fusion algorithms has been developed and presented in literature. The majority of them are based on multiresolution techniques of image processing.
The study of coastal zone and environmental change demands a diversity of accurate observation. Observation of processes on earth surface plays an important role in assessment of natural hazards in reducing their negative effect. Remote sensing applications is at very young stage but its results have been quite considerable. For decades, geologists have used satellite imagery to identify rocks and geological structures. Satellite imageries such as LANDSAT, SPOT and aerial photo are commonly used in geological mapping and exploration. LANDSAT has great advantage of large dimension, high spectral resolution and low price. However, in the study of coastal zone and environmental monitoring this type of imagery demonstrates its disadvantage of low resolution (25 meter of multispectral LANDSAT 7 and 15 meter of Panchromatic).
With the rapid increase in environmental preoccupation expressed by the international communities over the last decades the need for earth and environmental data has become more and more evident. The humanity has to face to global climate change, preservation of the biodiversity, desertification and natural hazards such as flood, typhoon, earthquakes, and landslides. With respect to all this subject, The use of high resolution satellite imagery contributes information, which is irreplaceable for the study of certain aspect of the problems. The use of high resolution satellite data in the study of coastal zone has been continuously developed since the launch of SPOT. The first generation of SPOT satellite with a 20 meter resolution in multispectral mode and 10 meter in panchromatic mode provides with large information for natural resources assessment and coastal zone monitoring. The combination of high resolution SPOT imagery and ground observation has contributed to an increasing understanding and management of natural resources and environment. There is important development with the new generation of Indian satellite images IRS-C with 5 meter in Panchromatic. This high resolution satisfies the demand for mapping at 1/50.000 scale in many countries. Spatial resolution of multispectral IRS-C limits at 20 meter as SPOT MS. The disadvantage of IRS-C is its price and limited cover. IKONOS imagery is a commercial product from 2000. High spatial resolution of IKONOS with a 4-meter in multispectral mode and 1 meter in panchromatic mode is excellent for earth observation and environmental monitoring. This imagery meets almost request of remote sensing application. However, IKONOS imagery has also its disadvantage. We know already that the repeat of observation is inverse proportional to the finest of the imagery. That is why one can not get easily the IKONOS imagery. It is more difficult for getting this imagery in tropical countries like ViÖt Nam due to almost permanent cover of cloud. In other side, the data are very expensive due to limited dimension and price. If we can make successfully the fusion of aerial photo and other satellite imagery such as LANDSAT and SPOT, we can have multispectral satellite imagery with high resolution and low price. This will open the great capacities in remote sensing application for coastal zone observation, urban mapping, environmental monitoring and natural hazard assessment. In this paper, we present for the first time our result of fusion between aerial photo and SPOT imagery and discuss about the possibility of integration of multisensor data on the study of coastal zone, such as the fusion of RADARSAT imagery with SPOT. Our procedure can be applied for the fusion of LANDSAT imagery with NOAA imagery for environmental monitoring. In Hải Phòng area, we use also the fusion technique to integrate SPOT with RADARSAT imagery
II. FUSION TECHNIQUES
Fusion techniques are a new advance in remote sensing. The fusion of multisensor and multiresolution satellite imagery is an effective means of exploiting the complementary nature of various images types. With the fusion of mutispectral imagery, higher spatial resolution panchromatic imagery is fused with lower resolution multispectral imagery. The fusion creates a product having the spatial resolution of panchromatic image and the characteristics of mutispectral image. The spectral characteristics are useful for identifying features such as trees, houses, soil, water etc. If the RADAR interpherometry is a powerful tool for determining fault displacements and terrain movements of landslides with the precision of some centimeters (Masonnet et al., 1993), fusion techniques permit one to combine an image of high spatial resolution with a multispectral image but of low spatial resolution for obtaining an other image in the same time rich in spectral and high spatial resolution. Various fusion techniques are used currently such as Intensity - Hue - Saturation transform (IHS), principal components analysis (PCA), high-pass filter (HPF) and wavelet transform. Basing on principle of fusion techniques, one can classify into 2 categories of fusion. The first one consists of fusion methods taking into consideration all channels in the fusion process such as IHS and PCA methods. The last one treats individually the information for each channel such as HPF and Wavelet transform methods [3, 10, 16]. The simplest method for fusing imagery is through arithmetic operations such as addition and multiplication. This operation can be used to merge varying degree. For example, in integrating a SPOT panchromatic image to the Thematic Mapper scene, we can specify weighting of 50 percent for each scene or weight the SPOT scene more important by using a 70 percent split. The used Color Normalized algorithm is one of the more sophisticated arithmetic techniques. The Color Normalized transform is a method for combining data such as SPOT panchromatic and LANDSAT imagery. The fusion of these two datum sets produces an image with the distinct spatial features provided by the TM, while the SPOT imagery provides increased spatial resolution [7]. The Color Normalized transform separates the spectral space into Hue and brightness components. The transform multiplies each of the three multispectral bands by the higher resolution panchromatic imagery and these resulting values are each normalized by dividing by the sum of the three multispectral bands. The method of Intensity - Hue - Saturation transform (IHS) is simple technique but it gives us effective result. From 3 parameters Red, Green, Blue of a multispectral image, one can transfer to Intensity - Hue -Saturation (IHS). These values IHS are characterized for the given multispectral imagery. The transform defined three separate, orthogonal attributes if intensity, Hue and saturation. The IHS can be represented as a cylinder. The intensity, which is the brightness or total energy of the image, is defined by vertical axis. The Hue, which is the average wavelength of color, is defined by the circumferential angle of the cylinder and has a range from blue through green, yellow, red and purple. The saturation, which is the percentage of white light in the image, is defined by the radius of the cylinder. When performing IHS transform, an analysis can use any three multispectral bands although the transform works best if the data are highly correlated. For example, TM bands 3,2, and 1 can be merged with SPOT panchromatic data. If we replace Intensity in multispectral images by panchromatic image of higher resolution, we will obtain new value IHS characterized for a new multispectral image. Re-transferring from IHS to RGB, we will get new image conserving spectral but high spatial resolution as old panchromatic image [2, 6]. Another step that may required prior to fusion is enhancement of both images to be merged. Contrast enhancement of the image will generally produced results containing more useful information. Edge enhancement is also sometimes included as part of the procedure. The High pass filter method consists of three basic steps: 1/ performing a high pass filter on the high resolution panchromatic image to extract the spatial information from this images, 2/ performing a low pass filter on the multispectral image to extract the spectral information from this image, and 3/ calculating the weighted sum of the high and low pass filtered results to create a band of sharpened product [3]. Wavelet transform methods can be affected by various approaches [15, 16, 18, 19]. Almost all workers make the fusion of Panchromatic and multispectral SPOT images having the different of spatial resolution of 2:1 [3, 15, 16, 19].
III. RESULT
The main question is that can we make the fusion of panchromatic imagery with multispectral imagery having spatial resolution ratio more than ten times? One of most important problem in the fusion is making georeference of aerial photo and satellite image. Thank to our technique of registration from image to image, we can compare the centre of a pixel of aerial photo with a part of pixel of satellite image with a dimension larger from 10 to 20 times. Iterative relative control points refining them through correlation, deriving transform coefficient and resampling the multispectral data to the panchromatic coordinates as reference carry out the registration. The resampling methods is cubic convolution. Cubic convolution uses 16 pixels to approximate the sinc function using cubic polynomials to resample the image. The type of polynomial warping available in our study is 3rd degree.
We test to integrate aerial photo of 1992 with SPOT multispectral in 1994 in C¸t H¶i , H¶i Phßng. C¸t H¶i Island of H¶i Phßng City is located next to C¸t Bà National Natural Park and §×nh Vò industrial park. SPOT, LANDSAT, ERS, RADARSAT and IKONOS imagery is available in this area (Figures 1 and 2). This area is well known for shrimp culture.
Figure 1. Multispectral SPOT image in 14 December 1994 at C¸t H¶i, H¶i Phßng.
Figure 2. Multispectral IKONOS image in 2000 at C¸t H¶i, H¶i Phßng
Using topographic map and control points in the C¸t H¶i region, we make irstly georeference of aerial photo. We compare georeference of SPOT image with that of aerial photo by using 22 control points. The control points are measured by GPS with the accuracy of less than 1 meter.
Figure 3. Multispectral SPOT imageat Hoàng Ch©u, C¸t H¶i, H¶i Phßng, 14 December 1994
We try to find the control points at the boundary of aerial photo. This work permits to increase the accuracy at the center of image. For example, if the error for georeference is 1/2 pixel of SPOT image at the boundary, the error at two control points at two corner of images is 1 pixel of SPOT image. If the distance of two control points is of 4000 pixels, the error at the center of fusion image is only 1/2000. This simple technique permits to compare georeference of aerial photo and satellite image with very high accuracy. In our case, many control points are situated far from the area of fusion, in C¸t Bà Island and in §×nh Vò, H¶i Phßng. We use aerial photo as master image and SPOT image as slave one for geometric correction. In the C¸t H¶i coastal zone, the quality of SPOT MS imagery is intensively ameliorated, so one can map small road of the villages, the dikes, small salt marsh and shrimp basin. In other part, thank to the conservation of multispectral resolution, one can more understand the sedimentary transport in shallow and mangrove mapping. For setting of the advantage of fusion image, we compare the SPOT multispectral with the fusion image in Hoàng Ch©u Commune, southwest C¸t H¶i (Figures 3 and 4), we observe more detailedly the structure of spring, sand beach and houses in villages.
The land use and salt marsh can be mapped in detail. Due to the change of mangrove distribution between the SPOT multispectral and aerial photo, the western border of Hoàng Ch©u Commune is not fused in fusion image. Basing on the dimension of SPOT MS not to be fused, we determine the reduction of mangrove as about 20 m, if we compare the fusion image with SPOT MS in southern part of C¸t H¶i region.
Figure 4. Fusion between aerial photo of 1992 and Multispectral SPOT image of 1994 at Hoàng Ch©u, C¸t H¶i, H¶i Phßng
Although the quality of fusion image is much higher than SPOT MS, one can judge that the integration between aerial photo and SPOT in this part is less than in Hoàng Ch©u Commune. It can be explained that tidal area is much changed between aerial photo and SPOT MS. We compare the fusion image with SPOT in southern part of C¸t H¶i area. The distribution of vegetation cover is clear in fusion image. Comparing the fusion image with IKONOS image, we regconize that the quality of fusion imagery is compatible with the information from KONOS images. In certain case, we can observe more clearly in fusion image (Figure 5).
Figure 5. Comparison of IKONOS imagery and fusion imagery.
One can observe more detail in fusion image than in IKONOS image. Using the most simple classification, we show the advantage of fusion image for recognizing small object (Figures 6 and 7).
Figure 6. Isodata unsupervised classification of SPOT image at Hoàng Ch©u, C¸t H¶i
Figure 7. Isodata unsupervised classification of SPOT image at Hoàng Ch©u, C¸t H¶i
We present also example of the fusion of multisensor data. We test the fusion between RADARSAT image, high resolution (12.5m) with SPOT. This procedure has advantage to combine the advantage of texture and resolution of RADARSAT imagery with SPOT multispectral imagery. The fusion image give some more complementary information. In the same time it loses also some other informations (Figure 8). Analysis of aerial photos of 1952, 1993 completed by SPOT-HRV together with LANDSAT-TM satellite images allows mapping the erosion and accumulation areas and to highlight the relation between these coastal geomorphologic processes and the wave patterns.
For evaluation of the spectral quality, we consider the histogram of original SPOT image with that of fusion image for each band. We compare the fusion image with IKONOS image. We compare also some land-cover classes. The average different is less than 10 % for 3 bands. This result is compatible with the previous study on fusion by IHS technique.
In our study, we meet an important difficulty which is the different date of aerial photo (1992) and of SPOT image (1994). This is a main limit of fusion technique. It will be better if we can fuse two images of small difference in date. Our result is also limited on IHS technique which give the spectral conservation much less than the result of wavelet technique. This is the objective of our work in the near future.
Figure 8. Fusion between SPOT image of 1994 and RADARSAT image of 1996.
IV. CONCLUSION
It is the first time that we make the fusion of aerial photo with SPOT MS having the difference of resolution of more than 10 time. This approach permits one to have large application in environmental observation. This procedure can apply to fuse Modis images with LANDSAT. Our work is the contribution for the Project GIS and remote sensing for the study on geological hazards, granted by RÐseau de TÐlÐdÐtection, UREF. The authors thank also the support from Ministry of Sciences and Technology of ViÖt Nam for the project in cooperation with University of Liege "GIS and remote sensing for the study of coastal zone in ViÖt Nam, example in H¶i Phßng and Nam §Þnh".
REFERENCES