Dynamic region of interest extract method for JPEG2000 coding

Region of interest (ROI) coding has became a hot topic in the research of JPEG2000. By using the wavelet coefficients which produced in the process of discrete wavelet transform (DWT) in JPEG2000, this letter proposed a new method to get ROI dynamically. Based on scales of wavelet coefficients, this method uses hidden Markov Tree and Gauss mixture model (HMTGMM) to get the ROI. The training set is optimized to get more accurate classification. In order to satisfy this method, a new bitplane shift tactic is designed, called Partial Bitplanes Shift (PBShift), to instead of the standard method in JPEG2000. PBShift supports arbitrarily shaped ROI coding and allows that different wavelet subbands can have different ROI definitions. Our results demonstrate that these methods can get ROI with smooth boundary dynamically.