APPROACHES TO PREDICTION OF SPECKLE REMOVAL EFFICIENCY FOR DCT-BASED FILTER
Several approaches to prediction of despeckling efficiency for DCT-based filter are presented and compared. The approaches allow predicting standard quantitative criteria as improvement of PSNR (IPSNR) as well as criteria of visual quality for filtered images. We propose and analyze rather accurate automatic procedures of prediction that exploit moments of a statistical parameter calculated in 8x8 pixel blocks of a given noisy image under condition that speckle parameters (or number of looks) are a priori known or pre-estimated with a proper accuracy. It is also shown that the prediction approaches are applicable to images with different intensity of speckle. Prediction based on neural network specially trained for multiplicative noise is demonstrated to be the most accurate.
Techniques for Earth observation data acquisition, processing and interpretation