Statistical modeling of error resilient JPEG2000 decoding

We propose a method to avoid false positive segmentation symbol decoding at the end of an erroneous bit plane in JPEG2000 encoded images transmitted over error prone environments. Within the JPEG2000 standard, its segmentation symbol is a utility for error detection. If this symbol is not decoded at the end of a bit plane, error will be detected; and the mentioned bit plane and the remaining ones will be filled by zeros-Zero Filling (ZF) method. If this symbol is decoded in the expected place by mistake, the erroneous bit plane will be taken as the true bit plane. Here, a technique based on non-zero decoded coefficient distribution in bit planes, using generalized Gaussian distribution (GGD) model, is investigated to avoid wrong segmentation symbol detection. The resulting decoder will be robust against errors caused by wrong segmentation symbol detection.