DCT coefficient statistics in images corrupted by spatially correlated noise

The analysis task of image discrete cosine transform (DCT) coefficients in case of spatially correlated noise is considered. Several estimators of shape parameter of generalized Gaussian distribution based on the forth central moment as well as percentile coefficient of kurtosis are applied in the research. It is shown that spatially correlated type of noise considerably influences statistics of DCT coefficients both in each particular image block and for a given spatial frequency for all image blocks. Histogram analysis of shape parameter estimates shows that noise type can be distinguished based on the mode position and its value.