Tesis doctoral de Otoniel Mario López Granado
Image and video compression is of utmost importance in multimedia systems and applications because it drastically reduces bandwidth requirements for transmission and memory requirements for storage. Great efforts have been made to improve coding efficiency of wavelet-based image encoders, achieving in this way a reduction in the bandwidth or amount of memory needed to transmit or store a compressed image. Unfortunately, many of these coding optimizations involve higher complexity, requiring faster and more expensive processors. For example, the jpeg 2000 standard uses a large number of contexts and an iterative time-consuming optimization algorithm called post-compression rate distortion (pcrd) to improve coding efficiency. Other encoders like embedded conditional entropy coding of wavelet coefficients (ececow) achieve very good coding efficiency with the introduction of high-order context modeling, being the model formation a really slow process. Even bit-plane coding employed in many encoders like set partitioning in hierarchical trees (spiht) or subband-block hierarchical partitioning (sbhp) results in a slow coding process since an image is scanned several times, focusing on a different bit-plane in each pass, which in addition causes a high cache miss rate. The aforementioned encoders are designed to obtain the maximum performance in rate-distortion terms, but unfortunately other design parameters like complexity or memory resources are not considered as critical as the former ones. recently, there has been increasing interest in the design of very fast wavelet image encoders focused on applications (interactive real-time image & video applications, geographic information system (gis), etc.) And devices (digital cameras, mobile phones, personal digital assistant (pda), etc) where coding delay and/or available computing resources (working memory and power processing) are critical for proper operation. In that scenario, the data must be encoded as soon as possible to fit the application time restrictions using the scarce available resources in the system (memory and processing power). Basically, these encoders do not present any type of iterative method and each coefficient is encoded as soon as it is visited. This process results in the loss of signal to noise ratio (snr) scalability and precise rate control capabilities. They simply apply a constant quantization to all the wavelet coefficients, encoding the image at a constant and uniform quality, as it happened in the former jpeg standard, where only a quality parameter was available (and no rate control was performed). in this thesis, we propose several rate control algorithms for non-embedded encoders. These algorithms will predict the proper quantization values that lead to a final bit rate close to the target one. In particular, we propose several bit rate prediction methods with increasing complexity and accuracy. The proposed algorithms do not introduce great overhead in the coding process. Also, in this thesis, a sign coding stage is proposed in order to improve the coding efficiency of the non-embedded encoders. regarding video coding, a wide variety of video compression schemes have been reported in the literature. Most of them are based on the discrete cosine transform (dct) and motion estimation/compensation techniques. However, a lot of research interest was focused on developing video wavelet coders due to the great properties of wavelet transform. Most wavelet-based video encoding proposals are strongly based on inter-coding approaches, which require high-complexity encoder designs as counterpart to the excellent r/d performance benefits. Even so, some applications like professional video editing, digital cinema, video surveillance applications, multispectral satellite imaging, high quality video delivery, etc. Would rather use an intra-coding system that is able to reconstruct a specific frame of a video sequence as soon as possible and with high visual quality. So, the strength of an intra video coding system relies on the ability to efficiently exploit the spatial redundancy of each video sequence frame avoiding complexity in the design of the encoding/decoding engines. in this thesis, we propose a new lightweight and efficient intra video coder, called motion lower tree wavelet (m-ltw), based on the lower tree wavelet (ltw) algorithm with accurate rate control capability and scene changes detection. This intra video encoder is able to encode an international telecommunication union (itu) d1 size sequence in real time with good quality. finally, in this thesis, we will introduce a 3d-wavelet based video encoder in order to improve the coding efficiency (more compression rate at the same quality than the intra video encoder), by exploiting the intrinsic video temporal redundancy.
Datos académicos de la tesis doctoral «Fast and efficient coding tools for digital image and video signals«
- Título de la tesis: Fast and efficient coding tools for digital image and video signals
- Autor: Otoniel Mario López Granado
- Universidad: Miguel hernández de elche
- Fecha de lectura de la tesis: 05/05/2010
Dirección y tribunal
- Director de la tesis
- Manuel Perez Malumbres
- Tribunal
- Presidente del tribunal: pedro angel Cuenca castillo
- rouzbeh Razavi (vocal)
- vicente González ruiz (vocal)
- Carlos Miguel Tavares de araujo cesariny calafate (vocal)