Evaluation of Adaptive Filters for Speckle Reduction in RISAT-1 Data for Flood Mapping

Processing SAR images without reducing speckle would yield inaccurate and unreliable results. To overcome this problem, many researchers have worked on development of different speckle reduction techniques/filters. However, the choice of best speckle reduction filter is subjective to data and objectives of the study. So, this paper focuses on evaluating the performance of different speckle filters, which work on the principle of reducing speckle by statistically manipulating the value of target pixel considering the values of neighbouring pixels. Five filters comprising - Lee filter, Lee Sigma filter, Frost filter, Local Region filter and Gamma filter were examined for their efficiency in speckle reduction over RISAT-I (MRS) data. The process of choosing an optimal filter is a trade-off between reduction of speckle and loss of useful data. So, the terms of trade-off were defined first. Since the aim of speckle reduction was to utilize the resulting filtered image for flood mapping, the specifications to determine the efficiency of the filter was defined accordingly. The specified characteristics of the filtered images were measured by Mean Square Error, Signal to Noise Ratio, Speckle Suppression Index, Speckle Mean Preservation Index and comparison of the change in Mean and Standard Deviation and also by close visual examination. The RISAT-1 data was filtered using all the five filters in three window sizes – 3×3, 5×5 and 7×7. The impact of various filters was studied on the entire image as well as water bodies separately. Keeping in mind that an efficient filter should reduce maximum speckle while preserving features and minimal loss of useful data, it was concluded that a single pass of Frost (7×7) is the most suited filter for RISAT-1 data intended for flood mapping application.