This paper proposes a statistical distortion measure for evaluating lossy compression methods of satellite images. The introduction of an adequate distortion measure is essential not only for users of satellite images but also for designers of satellite image compression methods. If the distortion measure were given, the users could know if the compressed images are reliable for their quantitative analysis and the designer could optimize their methods. The proposed distortion measure is a local Mahalanobis distance that is based on the covariance matrix of noise components in the images. The noise components are estimated from the second order spatial derivatives within a local window around the concerned pixel. Under the assumption of smooth signal components, it is shown theoretically that the local Mahalanobis distances of noise components follow the x2 distribution. In addition to this property, the local Mahalanobis distance is invariant to linear transformation of the values of multispectral bands. Then, the satellite image compression using pyramid linking segmentation is proposed where the local Mahalanobis distance is used as a similarity measure for linking. The proposed lossy compression method shows a good performance with regard to the proposed distortion measure compared with the standard lossy image compression method (JPEG) in the case of satellite images with a large number of bands such as LANDSAT TM image.
JERS-1/OPS has a special function to take a pair of stereo images by nadir and forward viewings. Observation time gap of about 20 seconds is contained in two images. Moving objects, e.g., ships, aircrafts, or clouds change their positions in this short period. By detecting the slight displacement of these targets, their speed vectors can be obtained. Applying this principle for a pair of OPS images taken on September 9, 1992, a ship speed vectors map of Tokyo Bay was obtained. In this map, speed vectors of about 400 ships on the bay and two aircrafts just landing at Tokyo International Airport are included. As for cloud speed vectors, a measurement of cloud shadow displacement on the ground may give its better result.
This paper describes the C-band backscattering variations and the development of a numerical model of backscatter from lake ice growing on shallow tundra lakes in northern Alaska. This research was prompted by earlier observations of unusual X- and L-band backscatter variations in side-looking airborne radar (SLAR) data, and the recent availability of a well-calibrated time-series of SAR data and concurrent ground observations and measurements. The North Slope of Alaska is a large expanse of low-lying tundra with many shallow lakes which comprise more than 40% of the surface area. During the 1970s, in late winter, SLAR images of these lakes showed interesting variations in backscatter intensity : areas of low backscatter at lake margins were believed to represent ice that was frozen to the lake bed while areas of high backscatter represented floating ice that contained tubular bubbles which acted as forward scatterers. From September 1991 to April 1992, backscatter intensity variations from shallow tundra lakes near Barrow, NW Alaska, were studied using C-band SAR data from the ESA Remote-Sensing Satellite-1 (ERS-1). The SAR data were processed at the Alaska SAR Facility (ASF) and the backscattering coefficient was derived for a number of lakes. Field measurements in April 1992 confirmed that the highest values were associated with floating ice containing tubular bubbles, while the lowest backscatter values were associated with ice frozen to the lake bed. The ice frozen to the bottom of the lakes also contained tubular bubbles. Ice core measurements indicated that the lake ice comprised three layers : 1) a surface layer of granular ice with roughly spherical bubbles with radii smaller than a wavelength : 2) a layer of clear bubble-free ice, and ; 3) a bottom layer containing tubular bubbles resembling thin cylinders with lengths ranging from 15 mm to 91 mm and a radius much smaller than a wavelength. The number of tubular bubbles per square meter, i.e. bubble density or ice porosity, was quite variable. The backscatter model that has been developed comprises the following elements : 1) an ice layer of variable thickness ; 2) ice sub-layers with air inclusions of variable density, size and shape, including cylinders of finite length, prolate spheroids and spheres : 3) air-ice boundary is specular surface, while ice-water and icefrozen soil boundaries are specular or slightly rough. The model results confirm that backscatter is a sensitive function of the presence of an ice-water interface, with a roughly 30-times greater reflectivity than an ice-frozen soil interface. The model also shows that backscatter increases as the ice grows. The model has also been tested using bubble data derived from ice cores in April 1992. The modelled backscatter is compared with backscatter derived from ERS-1 SAR images obtained at the same time as the fieldwork. It is found that the modelled and actual backscatter have high correlation.
A method for the separation of kinetic temperature and spectral emissivity on the thermal infrared multispectral remote sensing data of the land surface was proposed. This method, named Mean-Maximum Difference (MMD) method, is based on an empirical relationship between the mean and the variation of spectral emissivity of the natural materials in the thermal infrared region. To give an additional equation to the under-determined problem of the thermal infrared remote sensing, the relationship between the mean (M) emissivity and the maximum difference (MD) which is the difference between maximum and minimum values of a spectrum was assumed to be linear. The evaluation using laboratory emissivity spectra of various natural materials indicated the assumption of the linear relationship between M and MD is reasonable. The effects of vegetation in a rock-dominated field and the reflection of the environmental radiation were studied using simple numerical simulations. The results suggested that although the in-situ pixel-averaged spectra may differ from laboratory spectra of rock samples, their Ms and MDs satisfy the same linear relationship as the laboratory spectra. The MMD method was applied to airborne thermal infrared multispectral remote sensing data, and emis-sivity spectra were obtained from the rhyolite-dominated surface, the coniferous forest, and their mixture. These spectra showed distinct features such as the emissivity trough due to Si-O bond stretching in rhyolite spectra, relatively high and flat spectra for the forest, and the gradual spactral change over the transect from the forest zone to the rhyolite zone. These were consistent with published laboratory spectra of the same kind of materials though some discrepancies probably due to residual errors of the atmospheric correction, vegetation, and the reflection of the environmental radiation were found.
Satellite infrared images enable us to recognize macroscopically ocean phenomena such as current, eddy, front and others. However, the recognition has still remained ambiguous in minute parts. Authors investigated a theoretical method to extract cold-streamers from NOAA/AVHRR images by using knowledge information for the cold-streamers. The cold-streamers are homogeneous cold water bands which were derived from cold water masses into warm core ring. The candidates of cold-streamer were extracted by image processing. They were examined by using the standard deviation of temperature and the fractal dimension of learning samples wich were extracted by an expert. Resultantly, the candidates were confirmed the cold-streamers. It is considered that this procedure is useful in combination with the procedure of the extraction of warm core ring eddy for the expert system by which some oceanic phenomena were detected.