Version 0.1

050_Conclusions/Conclusion.tex

@@ -0,0 +1,37 @@
+
+\chapter{\HIGHLIGHT{Conclusion and future work}}
+\label{Chapter:Conclusions}
+
+This chapter concludes this thesis, presenting some conclusions about this research work and some future research perspectives in the area of video transmission, error resilience and error concealment.
+
+%\vspace{10mm}
+\section{Conclusions}
+
+% This section presents a summary of the of the chapters presented before and a discussion of the contributions presented.
+%
+% Chapter~\ref{CHAP:3DVideo} presented a review on the main technologies for 3D video systems related with the formats for representation, coding methods and proposed storage and transmission systems. This chapter also describes the main features of the MVC extension of the H.264/MVC standard, which is the coding standard for the most 3D applications \mbox{(\emph{e.g.}, Blu-ray)}. Moreover, as this research focuses in transmission errors and error concealment, a review of some concealment methods proposed over the last years was presented in the Chapter~\ref{CHAP:Concealment}.
+%
+% The research work presented in this thesis leaded
+% %different contributions
+% to the development of an advanced 3D video transmission system. More specifically, our research focused on the study of the subjective impacts of transmission errors in 3D video, when the auxiliary view is subject to frame loss. These results were presented in Chapter~\ref{CHAP:SubjectiveEval}. The results demonstrated that the disparity is intrinsically related with the quality of the concealed 3D video, and that errors have higher effects in those sequences with higher disparity. Moreover, for longer error bursts, the results showed that it is better to switch into 2D visualisation rather than trying to restore the 3D effect by using frame concealment methods.
+%
+% The results presented in the Chapter~\ref{CHAP:SubjectiveEval} leaded to further experiments in order to evaluate the objective quality of the transmitted 3D video using different objective metrics. A transmission framework was presented in Chapter~\ref{CHAP:HierarchicalDVBT3D} and the quality evaluation was performed. The quality models, devised from the analysis of the quality results, are a novel contribution to the research community and also for the network planners. Through those models, estimated values of the quality using different metrics can be obtained from the packet loss ratio of the transmission channel. This can be easily measured at the receiver side. The quality models are able to cover different video quality features, as the signal-to-noise relation, the structural similarity and a more specific feature of the 3D video, the stereo perception.
+% The tools developed are part of the first contribution of this research work, which resulted in an efficient 3D receiver to cope with transmission errors in 3D video streams, delivered using the MPEG-2 TS encapsulation. Moreover, as part of the receiver side, an efficient 3D video decoder was achieved, that is able to detect and conceal the missing frames using different concealment techniques. This set of tools provides relevant knowledge, for further investigations in quality evaluation and error concealment.
+%
+% The second contribution of this research work was the development of efficient concealment methods, able to recover the motion field for missing frames in 3D video. Assuming the hierarchical transmission system, as presented in Chapter~\ref{CHAP:HierarchicalDVBT3D}, the proposed methods are an efficient alternative to the existing ones. They are mainly based on the information presented in the error-free view (\emph{i.e.}, the base view), since these information can provide more accurate clues to estimate the loss data in the auxiliary view, as described in Chapter~\ref{CHAP:PropFrameLossConc}. The presented results showed the effectiveness of the proposed techniques, when compared with conventional motion-based algorithms, achieving an average gain of $1.89\ dB$ in random frame loss and up to $6\ dB$ of gain in the objective quality of the reconstructed frame. Moreover, an improvement to the proposed method was devised, and a motion based decision approach was incorporated, combining both methods, thus allowing to improve the quality of the reconstructed frames up $2\ dB$.
+%
+% One may conclude that the research on alternative frame loss concealment techniques may bring new insights to achieve a more efficiently video decoding in the presence of error prone channels. Moreover, the subjective study presented in this dissertation provided relevant subjective factors that may be used in future developments of error concealment techniques. New representation formats for 3D video and novel coding approaches are being proposed, leading to new error concealment problems, and different ways to recovery the missing information, thus, the research work in this area should continue.
+
+
+\section{Future work}
+
+% In this research work, the technologies and the subjective quality impact of errors in 3D video were studied. Nevertheless, there are still a number of topics that require further investigation, and may lead to more efficient 3D video decoders, namely:
+% \begin{itemize}
+%   \item \textbf{3D video quality models extension to the channel level.} The 3D video quality models presented in Chapter~\ref{CHAP:HierarchicalDVBT3D} revealed to match with the experimental data. Thus, the combination with the hierarchical DVB-T channel  models could extend the proposed models to a new level, where the quality of the 3D video could be estimated from the channel conditions, instead of the packet loss. This would extend the relevance of the proposed models to the network planners and developers.
+%
+%   \item \textbf{Extension of the proposed method to a multi-view scenario.} The proposed concealment methods, described in Chapter~\ref{CHAP:PropFrameLossConc}, revealed good performance to conceal missing frames in 3D video (\emph{i.e.}, stereoscopic video). Therefore, the same techniques are expected to have similar behaviour under a multi-view scenario, where, besides the base view frames, other auxiliary view frames that may provide motion information clues to estimate the lost frames' motion field.
+%
+%   \item \textbf{Extend the concealment algorithm to MVD coding scenario.} Although the proposed concealment methods did not focus in the recovery of the depth information, they could be used to recover texture in multi-view plus depth video signals. In this case, besides disparity information, depth maps are available to extract relevant information, for example, to detect occlusions and errors in the disparity map. The performance of the proposed method could also be investigated to reconstruct the lost depth map.
+%
+%   \item \textbf{Performance evaluation of the concealment methods in a MDC.} The multi-description coding approach is being investigated over the last years, as a novel alternative for the emergent multi-path networks. The algorithms presented in the proposed method have great flexibility and could be applied under different coding scenarios. Moreover, since MDC introduces new possibilities to error concealment by using the redundant information in each description, such characteristic will be used to investigate how different types of redundancy can be jointly used with the proposed concealment techniques to enhance the quality of received views.
+% \end{itemize}