Region of interest determined by picture contents in JPEG 2000

The region-of-interest (ROI) coding in Joint Photographic Experts Group (JPEG) 2000 is one of the most important coding methods in progressive transmission. Especially in transmitting video frames on a limited bandwidth, the ROI coding has been known to produce good perceptual quality in the specified region. According to the part 1 of JPEG 2000, the Maxshift scheme is utilized in the ROI coding. This work proposes a technique to automatically detect the ROI mask during the compression of JPEG 2000. The discrete wavelet transform (DWT) coefficients during the compression would provide clues to analyze the picture content and create a ROI mask. During the bit plane coding, the DWT coefficients in the significant propagation pass of the Embedded Block Coding with Optimized Truncation (EBCOT) are evaluated to automatically generate the ROI mask by using three steps of sub-block classification, central point determination and mask generation. As compared to the conventional ROI coding using the fixed and pre-determined area, the proposed scheme can provide more clear edges in multiple object shapes. Additionally, the ROI mask is updated by the picture content. The proposed scheme can be easily realized with the JPEG 2000 at a low computational cost.