Texture Estimation in Sar Images: the Impact of Scale and Model Parameters

When it comes to remote sensing image classification, the use of SAR image data is often regarded as being cumbersome and error-prone. Compared with optical images, SAR images have a very high dynamics, are plagued by speckle phenomena, and their radiometric calibration is less accurate than that of optical images. Many researchers have attempted to overcome these difficulties by applying, for instance, multi-look or de-speckling techniques prior to classification; the former methods depend on the initial number of looks and also on the ratio between resolution and pixel spacing. These aspects motivated us to investigate the impact of sub-sampling and multi-looking strategies.