Prediction of DCT-based denoising efficiency for images corrupted by signal-dependent noise

Prediction of Despeckling Efficiency of DCT-Based Filters Applied to SAR Images

“…However, below we will analyze only single-parameter dependences and the corresponding predictors as simpler cases. The probability P 0.5ı has shown itself as informative parameter providing more accurate prediction than other similar probabilities (as, e.g., P 2ı , see data in [20,21]). Therefore, we prefer to rely upon P 0.5ı in our further studies.…”

“…It has been also shown [20] that there is no necessity to use all possible blocks of an analyzed image for obtaining local estimates of input parameter. It is possible to use nonoverlapping blocks if their total amount is not less than 300…500 or 500 blocks selected randomly.…”

“…Fourth, prediction is possible using different input parameters although accuracy of prediction considerably depends upon chosen parameter(s) [15,18]. Fifth, prediction can be performed not only for AWGN [15,18,19], but also for signal-dependent noise [20,21] under condition that parameters of signal-dependent noise are known in advance or pre-estimated [16,27,28]. Sixth, initial results demonstrating possibility of prediction for spatially correlated additive noise have been obtained recently [29].…”

“…Meanwhile, there are recently proposed approaches to prediction [15,[18][19][20][21] that are fast and based on another principle. It is supposed that there exists a rather strict dependence between one or several "input" parameters (that can be easily calculated for an analyzed image) and a parameter that characterizes denoising efficiency.…”

“…Recently, some steps towards designing prediction procedures have been done [15,[17][18][19][20][21]. In [17], it has been shown that for non-local denoising methods applied to remove additive white Gaussian noise (AWGN) it is possible to predict potential output mean square error (MSE) without having the corresponding reference image.…”