Automation of analysis for correlated noise in images based on DCT coefficient statistics

Knowledge of noise properties is important for many applications. It is often necessary to estimate characteristics of noise, in particular, spatially correlated noise automatically. For this purpose, it is desirable to have methods that allow obtaining preliminary estimates of spatial correlation degree. Here we show how this can be done by processing discrete cosine transform (DCT) coefficients calculated in 8×8 pixel image blocks. Moreover, we test the proposed method for synthetic aperture radar (SAR) images with different number of looks. The performed tests demonstrate that for TerraSAR-X data the noise correlation degree increases for larger number of looks. Besides, it is shown that the approach earlier tested for additive noise is applicable to pure multiplicative noise (speckle) and, in general to other kinds of signal-dependent noise.