A problem of texture preservation in MM-band radar images while their filtering is a task important for remote sensing data interpretation [1]. Unfortunately, in many cases of filter design and application the texture preservation property is not properly taken into account. This happens even in case of application of locally adaptive filters with hard switching where the image homogeneous regions (locally passive areas) are processed by some noise suppressing filters (NSF) and locally active regions (edge and detail neighborhoods) are processed by the so called detail preserving filters (DPF) [2]. The discrimination between them is done on basis of local activity indicator (LAI) analysis, i.e. their comparison to some preset threshold. The texture is usually referred to locally active areas and DPFs generally preserve it better than NSPs. However, as it will be proven later, different DPFs are worth applying to edge/detail neighborhoods and texture areas. So, a reasonable idea is to apply different filters for processing the edge/detail neighborhoods and the regions with texture within the advanced hard switching framework. Thus, it is desirable to discriminate the edge/detail neighborhoods from textural areas and to design three state hard switching filtering schemes.
展开▼