Efficient motion compensated prediction in video coding
Student thesis: Doctoral Thesis
Related Research Unit(s)
Modern video coding schemes achieve high compression efficiency by exploring the temporary redundancy between frames via motion compensated prediction. The most successful approach so far is block-based motion compensated prediction. In block-based motion compensated prediction, a block of pixels in reference frames is chosen as prediction candidate for the block in the current frame. The process to choose the best-matching is called block-matching and many different block-matching algorithms have been proposed. The prediction accuracy of the selected block-matching algorithm greatly affects the coding efficiency of motion compensated prediction. To enhance the efficiency of motion compensated prediction, many techniques have been proposed. Conventional fractional-pixel motion estimation researches focus on the use of quadratic equations to model the distortion surface. However, quadratic equations are not always solvable. On the other hand, the neighboring block prediction method does not perform well for complex motion sequences when the neighboring blocks are not correlated. A piecewise linear model is proposed in this work to represent the fractional-pixel error surface. This model has an intrinsic center-biased characteristic because it uses the higher distortion-declining rate of the neighbor integer-pixel distortions to define the slopes of two linear equations. Based on this model, a fast fractional-pixel motion estimation algorithm is developed which is fast and robust. Block-based motion compensated prediction assumes that objects move along the imaging plane with translational motions but projection of real world moving objects onto a two-dimensional imaging plane will not always result in pure translational objects motion. One prevailing motion type in videos is rotational motion, for example the rotation of a car wheel or the waving of a hand, etc. Moreover, many complex motions can be modeled by translational and rotational motion combined. In this work, translational and rotational motion compensated prediction is implemented by sub-sampling in the interpolated reference frame. This approach does not require the estimation of motion parameters and has the merits of easy implementation and low overhead. The interpolated frame used by translational and rotational motion compensated prediction is the same as that used by motion compensated prediction with fractional-pixel accuracy existing in most recent video coding standards. Experimental results found that up to 37 % of the blocks can be better predicted with rotational motion compensated prediction. The proposed method has the merits of easy implementation and low overhead since it needs to transmit one rotational angle parameter only. In multiview video coding, disparity-compensated prediction exploits the correlation among different views. A common approach is to use block-based motion-compensated prediction to predict disparity effect among different views. However, some regions in different views may have various deformations due to non-constant depth. It is found that horizontal scaling and shearing deformations are common among images of horizontally aligned views. Using similar special sub-sampling in interpolated reference frame technique, horizontal scaling and shearing based disparity-compensated prediction is proposed which does not require affine parameters estimation and additional frame buffers. Finally, a new motion compensated prediction method using superimposed inter-frame signals to achieve higher prediction accuracy is proposed. In theory, multi-hypothesis motion compensated prediction can enhance the prediction quality of motion compensated prediction. Traditional multi-hypothesis motion compensated prediction methods use fixed weightings for the linear combination of the multiple signal sources which may not be optimum. Moreover, multi-hypothesis motion compensated prediction requires the transmission of more than one motion vector, which will increase the side information to be transmitted. It is discovered that by using estimated distortion ratio, a weighting pair can be estimated adaptively for the linear combination of two signal blocks to form a prediction block with a lower distortion. The proposed method has better prediction accuracy than conventional motion compensated prediction and yet does not require the transmission of additional side information.
- Digital video, Coding theory, Video compression