Region of interest determined by perceptual-quality and rate-distortion optimization in JPEG 2000

The region-of-interest (ROI) coding of Joint Photographic Experts Group (JPEG) 2000 is an important option to encode pictures on a limited bit rate. This work presents a scheme to automatically generate the ROI mask during the encoding process of JPEG 2000. In the proposed scheme, the embedded block coding with optimized truncation (EBCOT) is utilized to provide information of the significant states, rates and distortions at bit-planes for analyzing picture contents, and then determining an appropriate ROI mask that is optimized by the perceptual quality and rate-distortion function. During the bit-plane coding, the coefficients of discrete wavelet transform (DWT) in the magnitude refinement pass of the EBCOT are grouped into the interested or uninterested ones by using their pass scanning results from perceptual point of view. Furthermore, the rate-distortion function is applied to automatically determine the optimized ROI mask from the probable masks that are formed by the interested sub-blocks, according to the required bit rate. As compared to the scheme using the fixed-square or fixed-region ROI mask, the proposed scheme can provide clearer object contours and better image quality. Additionally, the proposed ROI determination scheme can be easily realized and embedded in JPEG 2000 encoding process.