Selecting the don't care bits in JPEG2000 ROI coding
Visual Communications and Image Processing 2004
At the time of this writing, JPEG2000 is the most recent standard for still image compression. One of the important features of the standard is region-of-interest (ROI) coding, which allows user-defined parts of an image to be coded with higher quality than parts in the "background" (BG). In fact, the ROI feature enables a general non-uniform distribution of quality for different parts of an image. There are two main types of applications for ROI coding. In many applications such as in medical
... mages, a high quality or even lossless quality is needed for only a small part of an image. In these cases, with using ROI coding technique, there is no need to code the whole image with high quality. The image background is coded with moderate quality and as a result a higher compression ratio is achieved. The second type ui appii~aiiur~ is ill prvgressive irarisrriission over iow-capaciiy cnanneis. in inese cases, the ROI is transmitted before the background, and at each point during the transmission, the quality of the ROI is better than the BG. Therefore, once the desired image fidelity is provided at the receiver, transmission can be terminated. Also the relative quality between the ROI and BG regions can be selected during the coding process. The JPEG2000 standard only exists for the decoder side. Any algorithm can be used at the encoder as long as it is compatible with the standard's decoder. In this thesis, we propose a modification for the current JPEG2000 encoders in order to improve the compression performance of the ROI mode of the standard. ROI coding is accomplished in JPEG 2000 by de-emphasizing the wavelet coefficients associated with the non-ROI regions of the image. A number of extra bits appear below the least iii significant bit of the ROI samples after the shifting process. These bits need to be coded at the time of bit-plane coding, but they are discarded by the decoder. The usual procedure is for the current encoders to set these "don't care" bits to zero. In this thesis, we examine the possible strategies for setting these "don't care" bits and then propose a method that exploits the state of the JPEG2000 entropy coder to set the values of these bits in a more intelligent way. The method has been observed to reduce the number of bits required to represent ROI code-block by up to 7%, with the bit-stream remaining JPEG2000 compliant.