![]() Geometric DistortionsThese types of distortions are induced by variations of the elevation in the surface, or changes in the platform atitude (velocity, direction and altitude). The variations of the surface elevation give rise to distortion known as foreshortening, layover and shadowing. The layover occurs when the top of a target is imaged before the base, causing an inversion of the terrain, with the high parts mapped as low, and vice-versa. This effect is always more intense with small incidence angles, as it is the case of orbital systems in general. The figure below shows the layover effect, characterized by several strips of white color in the entire scene. Figure - Image with layover effect. Foreshortening occurs when the imaged area has a pronounced relief. In this case, the slopes facing the nadir will be shortened. The following figures show graphically the distortions of foreshortening, layover and shadowing. Shadowing: the radar shadow is an area of no data or total black. Fig. - Foreshortening (1), layover (2) and shadowing (3). The corrections of these effects require additional processing, as they need information about the Digital Terrain Model (DTM).
Slant to Ground Range ConversionAnother type of geometrical distortion is due to the radar side looking geometry. The side looking generates an image with a projection inclined in relation to the ground, causing a compression of the image. This compression varies along the imaged area. The pixels nearest to nadir will be more compressed. The conversion of the inclined projection to the ground projection is called slant to ground range conversion. The slant image is related to the acquisition mode in side looking radars. The following figures show how the data acquisition process is performed. Figure - Sampling the received echo in intervals Ta With the sampling process, the information contained in each interval Ta does not come from the same area size for samples situated in near range than that situated at far range, due to the variation of the incidence angle
Figure - Slant and Ground Range Images The formed image is denominate slant range image. This image has a geometrical distortion, since the SR samples equally spaced in the imaged area are not equally spaced in the ground, GR. In order to register and geocode this image, the ground samples should be equally spaced, which requires the conversion of the slant range image to the ground range image. The conversion consists in projecting the samples (pixels) in the ground range and then resampling them with a uniform spacing. The conversion uses parameters related to SAR geometry such as flight altitude, minimum distance (distance of the sensor to the first pixel), minimum time (registered time from the sensor to the first pixel). These parameters are, in general, presented in the selected image header. In case they aren't, the fields height and minimum slant range should be filled in, or the fields minimum incidence angle or minimum time. Any of the last three parameters are enough for doing the conversion. Other information which should be considered is the right or left looking position, which can be identified by analyzing the shadows in the image caused by the SAR side looking. For resampling the slant image in order to obtain a uniform sampling in the ground, three types of interpolation can be used:
The relation between slant range resolution,
The ideal conversion is the one that takes into account the digital terrain model (DTM), allowing the correction of the distortions caused by the layover, shadowing and foreshortening. The DTM of the corresponding image is not always available. Airborne images, from mountainous regions, are in general converted to ground range assuming flat earth. In this type of image, the incidence angle is high, due to the low platform altitude; then, the layover effect almost does not exist but, if the region is mountainous, shadowing problems will be present. The figure below shows a slant range image (a) of Tapajós River, Brazil, gathered from SAR-580 system during the SAREX-1992 mission, and its corresponding ground range image (b). It can be seen that the right side of image (a), near range, is more compressed than that of image (b), due to the non-uniform sampling of the terrain.
(a) (far range) -------------------------------------- (near range)
(b) Figure - Slant range image (a) and its corresponding ground range image (b)
See Slant to Ground Range Conversion SAR Image RegistrationThe image registration is a process of superposing two or more images acquired from the same geographic region, making one image to be perfectly superposed to the other. The registration is an essential stage when one needs an integration of data obtained from different sensors (images/sensor fusion), temporal image analysis (temporal registration or change detection) etc In remote sensing, there exist a large number of natural resources sensors, with different geometric and radiometric characteristics. The combination of the images may improve the information extraction process. The success of SAR images registration depends on the degree of similarity between the images, such that non-similar images are difficult to register with good precision. The degree of similarity depends on the terrain topography, radar illumination angle, geometric resolution and Speckle noise. When the images are from a flat terrain, the geometric differences between them can be systematically removed, being possible to obtain a registration with high precision. When the images are from a mountainous terrain, the difficulties increase, especially if the images are obtained with different look angles since, in this case, the edges in the images change from one image to another, mainly due to the differences caused by shadowing and layover. Images belonging to ascending and descending orbits usually have differences in the incidence angles, and probably the registration will not have a good precision, especially on mountainous regions. When doing a manual registration between SAR images, the difficulties appear in the location of the control points , mainly due to the Speckle noise. In order to minimize this problem, the speckle noise should be reduced by using suitable filters that preserve the edges in the images, facilitating the location of line intersections that are natural candidates for control points. In the automatic registration, the most used method is the "area based registration". The method considers the correlation of small windows extracted from the images. The algorithm consists in shifting one window with respect to the other, and computing the correlation coefficient between them. When the correlation coefficient is maximum, the windows are registered; in this way, the displacement between the images in the window area can be known. By using several windows spread in the entire area contained in the images, it is possible to create a distortion model between the images. From this model, polynomials for registering the images are generated. The automatic registration can not always be applied to the entire area of the images, since there might exist regions of low similarity between the images, where a high correlation coefficient is not achieved. Registration between SAR and Optical ImagesThe different acquisition geometries of SAR and optical images, the different types of illumination (optical - passive, Sun; SAR - active, microwave) and the different spectral bands used, causes a low degree of similarity between the images from these two sensors, in the geometry as well as in the radiometry. The problem in registering these two types of images is that the structures in the scene have different returns for two types of sensors. The edges in the images, in spite of having problems of different returns, make possible the registration of these images. In the manual registration, the Speckle noise of the SAR image should be appropriated filtered, in such a way that the edges are preserved, for a better location of the control points. In the automatic registration, the edges correlation method has been used with success for some type of images. See the necessary procedures for image registration.
![]() |