| Summary: | 碩士 === 國立東華大學 === 電機工程研究所 === 89 === The ability to give higher priority to Region-of-Interest(ROI) is the emerging functionality for nowadays image/video coding. In this thesis, two different bit allocation strategies are proposed for Region-of-Interest(ROI) still image and video coding. Some improved issues are also proposed.
For still image coding, we incorporate multiple ROI coding and data embedded functionalities into SPIHT codec with reversible Two-Ten integer wavelet transform, and propose two improvements of SPIHT algorithm and utilize context-based adaptive arithmetic coder to further enhance coding efficiency. Each ROI has individual ellipse or rectangle shape model and possesses own priority(reconstruction speed). The priority decides the order of processing for the coefficients at encoder and decoder. In addition, we propose a method called OENA to embed data. The data can be the meaningful image or the useful text for some special applications such as digital camera and digital library. We implement a remote monitor system to integrate all techniques of this portion. In this system, the luminance is compressed by an improved SPIHT codec, and the chrominance is embedded into the bitstream by OENA。The guarder can decide to use ROI or not, and the bit-rates to transmit to network.
For video coding system, a region-based video codec based on the H.263+ video standard is set up. The associated rate control scheme is proposed for ROI coding through low bit-rate channels. The ROI can be defined statically or dynamically. A static ROI should be shaped to include the fixed space. The region of central rectangle is suitable for static ROI because the central region is usually more important than the surrounding region. The dynamic ROI is defined by a simple, fast, accurate face detection method that is used to locate the macroblocks that contain facial area in real time. We adjust the distortion weight parameter at macroblock layer by the rate control of TMN8 to ensure that the ROI gets better quality and more bits are allocated. From extensive experiment results, the proposed method can significantly improve subjective quality at facial region. It is very suitable for the applications of video conferencing and remote control.
|
|---|
