Approaches to prediction of speckle removal efficiency for DCT-based filter

Remote sensing (RS) from airborne and spaceborne carriers has found numerous applications [24]. Radar RS is applied alongside with other types of imaging sys tems and it provides certain benefits compared to oth er RS systems as ability to work during day and night and possibility to operate in bad weather conditions [15]. However, acquired radar images formed by mod ern synthetic aperture radars (SARs) suffer from a noise like phenomenon called speckle that appears due to coherent imaging mode and is the most intensive for single look SAR images [15, 26]. There are several ways to cope with speckle [15, 26]. One way is to form multi look images that, unfortunately, leads to worse spatial resolution of registered images. Besides, for a limited number of looks (e.g., two or three) speckle can be still intensive and annoying. Another way is to perform despeckling (filtering, denoising) where numerous methods exist (see [8, 9, 26] and references therein). However, such despeckling, alongside with efficient noise removal in image homogeneous regions, might smear edges/details and destroy texture features which is undesirable. These undesirable effects appear them selves for complex structure images containing a lot of fine details and textures. Then sometimes it becomes undesirable to carry out denoising or, at least, one has to perform filtering more carefully than usually. Thus, it is desirable to predict efficiency of despeckling be fore starting this operation of SAR image processing [3]. One problem is that despeckling efficiency depends upon many factors [8, 9, 26]. They are properties (complexity) of an image to be processed, noise statistical and spatial correlation characteristics, UDC 004.021