MMSE based seed selection in IDAN speckle filter with point target preservation

Abstract

Intensity Driven Adaptive Neighborhood (IDAN) filter, is an adaptive filtering technique where a neighbourhood is formed using region growing method around each pixel. Pixels belonging to an adaptive intensity driven region are more likely to respect the local stationarity hypothesis than pixels belonging to a squared fixed size window. The advantage of the IDAN algorithm as compared to the other filters, is that the selected pixels are not required to lie within a fixed moving window. Fixed neighbourhood speckle reducing techniques are less effective for variations in feature size and shape. In IDAN filter, just like the original Lee Sigma filter, bias of underestimation has been problematic. In this work, we propose a new polarimetric SAR speckle filtering algorithm, based on the IDAN filter, which addresses the problem of bias. In the proposed filter, bias problem associated with IDAN filter is minimized by selecting Minimum Mean Square Error (MMSE) based estimated seed value and strong reflective targets (point targets) are preserved by statistical technique as used in the Improved Sigma filter. The proposed filter has been evaluated by using fully polarimetric air-borne AIRSAR L and C-band data and ALOS-PALSAR L-band and RADARSAT-2 Cband space-borne data. Well known indices are used for the quantitative and qualitative performance evaluation. It is found that, the proposed filter shows reduction in bias as compared to IDAN filter, while having improved point target and edge preservation ability.