An adaptive three-dimensional DCT compression based on motion analysis

Furht, Borko; Gustafson, Ken; Huang, Haihong; Marques, Oge

doi:10.1145/952532.952683

Cited by 9 publications

(1 citation statement)

References 5 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The main drawback of a 3-D DCT based codec is the use of a fixed-length transform regardless of the level of motion activity. To solve this problem, variable length 3-D DCT schemes with multiple thresholds have been proposed [11][12][13][14]. These techniques apply variable transforms based on certain empirical thresholds.…”

Section: Energy Concentration and Coefficients Scanningmentioning

confidence: 99%

Contribuições em codificação de imagens e vídeo

Testoni¹

View full text Add to dashboard Cite

The image and video coding community has often been working on new advances that go beyond traditional image and video architectures. This work is a set of contributions to various topics that have received increasing attention from researchers in the community, namely, scalable coding, low-complexity coding for portable devices, multiview video coding and run-time adaptive coding.The first contribution studies the performance of three fast block-based 3-D transforms in a low complexity video codec. The codec has received the name Fast Embedded Video Codec (FEVC). New implementation methods and scanning orders are proposed for the transforms.The 3-D coefficients are encoded bit-plane by bit-plane by entropy coders, producing a fully embedded output bitstream. All implementation is performed using 16-bit integer arithmetic.Only additions and bit shifts are necessary, thus lowering computational complexity. Even with these constraints, reasonable rate versus distortion performance can be achieved and the encoding time is significantly smaller (around 160 times) when compared to the H.264/AVC standard.The second contribution is the optimization of a recent approach proposed for multiview video coding in videoconferencing applications or other similar unicast-like applications. The target scenario in this approach is providing realistic 3-D video with free viewpoint video at good compression rates. To achieve such an objective, weights are computed for each view and mapped into quantization parameters. In this work, the previously proposed ad-hoc mapping between weights and quantization parameters is shown to be quasi-optimum for a Gaussian source and an optimum mapping is derived for a typical video source.The third contribution exploits several strategies for adaptive scanning of transform coefficients in the JPEG XR standard. The original global adaptive scanning order applied in JPEG XR is compared with the localized and hybrid scanning methods proposed in this work. These new v orders do not require changes in either the other coding and decoding stages or in the bitstream definition.The fourth and last contribution proposes an hierarchical signal dependent block-based transform. Hierarchical transforms usually exploit the residual cross-level information at the entropy coding step, but not at the transform step. The transform proposed in this work is an energy compaction technique that can also exploit these cross-resolution-level structural similarities. The core idea of the technique is to include in the hierarchical transform a number of adaptive basis functions derived from the lower resolution of the signal. A full image codec is developed in order to measure the performance of the new transform and the obtained results are discussed in this work.

show abstract

Section: Energy Concentration and Coefficients Scanningmentioning

confidence: 99%