SAR Image Processing: Non-Statistical, Statistical, and Frequency-Domain Filtering Methods

Non-statistical filtering methods include the Mean Filter, Median Filter, and K-Nearest Neighbors (KNN) Filter. The Mean Filter operates by replacing each pixel value with the average of its neighborhood, effectively reducing noise through simple convolution operations. The Median Filter utilizes a sliding window to select the median value, providing robust noise suppression while preserving edges. The KNN Filter leverages machine learning principles by averaging the k closest pixel values based on distance metrics. Additionally, commonly used statistical filtering methods comprise the Lee Local Statistics Filter and Maximum A Posteriori (MAP) Filter. The Lee Filter adapts to local statistics by modeling speckle noise characteristics, requiring variance calculation within sliding windows. The MAP Filter employs Bayesian estimation to minimize error probability, implementing optimization algorithms for noise reduction. Another approach involves frequency-domain methods primarily utilizing wavelet transforms. Among these, the wavelet soft-thresholding method proposed by D.L. Doholo et al. stands as a prominent technique. This method applies thresholding to wavelet coefficients, where small coefficients (likely noise) are shrunk toward zero while preserving significant signal components through multi-resolution analysis.