Efficient computation of visual saliency regions has been a research problem in the recent past, but in the case of 3D content no definite solutions exist. This paper presents a computational method to determine saliency regions in 3D video, based on fusion of three feature maps containing perceptually relevant information from spatial, temporal and depth dimensions. The proposed method follows a bottom-up approach to predict the 3D regions where observers tend to hold their gaze for longer periods. Fusion of the feature maps is combined with a center-bias weighting function to determine 3D visual saliency map. For validation and performance evaluation, a publicly available database of 3D video sequences and corresponding fixation density maps was used as ground-truth. The experimental results show that the proposed method achieves better performance than other state-of-art models.

The increasing rate of creation and use of 3D video content leads to a pressing need for methods capable of lowering the cost of 3D video searching, browsing and indexing operations, with improved content selection performance. Video summarisation methods specifically tailored for 3D video content fulfil these requirements. This paper presents a review of the state-of-the-art of a crucial component of 3D video summarisation algorithms: the key-frame extraction methods. The methods reviewed cover 3D video key-frame extraction as well as shot boundary detection methods specific for use in 3D video. The performance metrics used to evaluate the key-frame extraction methods and the summaries derived from those key-frames are presented and discussed. The applications of these methods are also presented and discussed, followed by an exposition about current research challenges on 3D video summarisation methods.

This paper proposes an algorithm for automatic detection of 3D video shots with different perceptual features. The proposed algorithm is able to identify distinct three-dimensional visual scenes by detecting 3D video shot boundaries based on clustering of depth-temporal features. A combination of texture variation along the temporal dimension and depth variance is used by K-means clustering to find the stereo frames which comprised the 3D scene boundaries. An important characteristic of the proposed algorithm in comparison with others published in the literature for temporal segmentation of classic 2D video is that no thresholds are used in the decision processes neither training data sets. The experimental results show that the proposed method is capable of achieving high recall (e.g., 0.95) and precision rate (e.g., 1.0) in video sequences with both sharp and smooth 3D scene transitions.

This paper proposes a robust video streaming scheme for priority networks with path diversity, based on a combined approach of multiple description coding (MDC) with optimal picture classification into two priorities. A binary classification algorithm is proposed to define high (HP) and low (LP) priority network abstraction layer units (NALU), which in turn define the packet priorities. An optimisation algorithm is used to find HP pictures, based on dynamic programming and relying on minimisation of the packet loss concealment distortion. The paper shows that the proposed algorithm is able to effectively improve the decoded video without increasing the MDC stream redundancy. The overall performance evaluation, carried out by simulating MDC video streaming over lossy networks with path diversity, demonstrates that the proposed algorithm yields higher video quality for a wide range of packet loss rates (PLR). Comparing with no-priority MDC video streaming schemes, the simulation results show that the proposed algorithm can improve the average PSNR results in 0.7-3.2dB for packet loss rates between 3% and 15%.

A method to efficiently encode an arbitrary video summary with temporal scalability and dynamic Group of Pictures (GOP) structures is proposed in this paper. The video summary is encoded as the base layer of a Scalable Video Coding (SVC) bitstream using a novel approach, which matches the summary frames available in temporal segments onto corresponding dynamic GOP structures. An algorithm is devised to compute variable GOP sizes along with an efficient independent prediction structure for the summary. The results show that the proposed method can be used to encode arbitrary video summaries with increased efficiency in the temporal base layer and negligible loss of R-D performance in the whole scalable sequence.