Joan Bartrina Rapesta
This work is focused on Region Of Interest (ROI) coding methods for the JPEG2000 image compression standard, ROI coding is a feature provided by some modern image coding systems that allows prioritize specific ROIs over the rest of the image (the background).
The main idea behind ROI coding is to increase the priority of the ROI within the encoding process, generating a compressed codestream that, decoded at increasing bitrates, recovers the ROI first and with a higher quality than the background.
JPEG2000 provides ROI coding through two mechanisms: either modifying wavelet coefficients, or using rate-distortion optimization techniques. Although ROI coding meth- ods based on the modification of wavelet coefficients provide an excellent accuracy to delimit the ROI area (referred to as fine-grain accuracy), they penalize significantly the coding efficiency. On the other hand, methods based on rate-distortion optimization just modify the distortion computation avoiding to penalize the coding efficiency but, so far, have not been able to achieve the intended fine-grain accuracy.
This thesis introduces two ROI coding methods that enhance the fine-grain accuracy of ROI methods using rate-distortion optimization techniques, yielding an accuracy compara- ble to that provided by methods modifying wavelet coefficients. We propose two methods named Subblock and Weighted. They do not penalize the coding efficiency and can be easily incorporated in any implementation. Furthermore, they maintain JPEG2000 compliance.
The main difference between Subblock and Weighted methods is how they determine the distortion of the ROI. Subblock modifies the distortion computation at the coefficient level, whereas the Weighted modifies the distortion computation at a codeblock level. These two new methods have been compared with the Implicit method. The Implicit ROI coding method was the first method that use rate-distortion optimization techniques to prioritize the ROIs.
Experimental results suggest that Subblock and Weighted outperform the Implicit, both for the ROI and the background due to the better adjustment of the distortion. When ap- plied to multicomponent images, experimental results suggest that Subblock and Weighted methods improve the Implicit coding performance. The computational costs of both the Subblock and the Weighted is negligible.