Preferential Coding for Mobile Multimedia Services

Abstract: Different parts of source encoded multimedia streams such as those associated with standard image or video formats possess different levels of importance with respect to their contribution to the quality of the reconstructed image or video. This unequal importance among data within a codestream gives rise to preferential treatment of the more significant parts of the codestream compared to the less important parts. Similarly, visual information offered by certain regions of an image or video may attract viewer’s attention more than other parts of the viewing area. As a consequence, preferential treatment of important data and information can play a vital role in mobile multimedia services in order to preserve satisfactory quality of service under the harsh conditions of a band-limited, error-prone wireless channel. In this thesis, we therefore, investigate how preferential coding can be used to protect multimedia services more efficiently against transmission errors. For this purpose, an error sensitivity analysis of the specific application is utilized as a basis to design efficient unequal error protection (UEP) schemes. The performance of the proposed preferential coding schemes is evaluated using objective perceptual quality metrics in order to account for the fact that humans are the ultimate judges of service quality. The thesis is divided into four parts. In the first part, region of interest (ROI) identification, coding and advantages of ROI coding for different applications are investigated. In addition, a framework is proposed for using ROI coding in wireless imaging. The second part analyses the error sensitivity of wireless JPEG2000 (JPWL) in terms of perceptual quality metrics. It is also shown that using reduced- reference perceptual quality metrics as error sensitivity descriptor (ESD) in JPWL increases the effectiveness of ESD. Specifically, this type of metrics correlates well with subjective quality assessment and provides an additional estimate for the obtained quality. In the third part, two UEP schemes for JPWL are proposed and compared with equal error protection (EEP). Their performance is evaluated in terms of perceptual quality metrics such as structural similarity index and the visual information fidelity criterion of the reconstructed image and their benefit over EEP is revealed. Finally, in the fourth part, a framework for optimized preferential coding of ROI based images and videos is proposed. Specifically, a dynamic programming algorithm for optimal parity distribution is provided.