Grupo de Tratamiento de Imágenes




News and Events 

"Design and Optimization of Protection Strategies Based on the Pro-MPEG COP3 Codes for Time-sensitive Multimedia Streams" 

César Díaz

E.T.S. Ing. Telecomunicación, Universidad Politécnica de Madrid, Jun 2017, "Cum Laude".

Ph.D. thesis Director: Julián Cabrera Quesada.

In the recent years, an ever-increasing portion of the services related to video data transmission are supplied through IP networks, as they present numerous advantageous characteristics: ubiquity, easy service integration and synchronization, a far higher possibility for interactivity and many more. However, IP networks are not particularly well suited for video distribution. Thus, QoS management mechanisms are commonly employed to increase their reliability, particularly in server-driven, time-sensitive transmission scenarios. In these scenarios, the Pro-MPEG COP3 codes are commonly used, due to their capability to deal with packet loss bursts and to their very low complexity. However, their performance decreases with the packet loss rate. On the other hand, protection strategies that are aware of the uneven relevance of the different packets in the video data stream typically provide better performance, since they are capable of better distributing and using the available protection resources. However, their complexity usually exceeds that of the Pro-MPEG COP3 codes.

This thesis examines strategies to improve the performance of the Pro-MPEG COP3 codes without compromising their simplicity and preserving their encoding and decoding procedures. These strategies aim at a generalization and optimization of these codes, through, among other ways, enabling that they can be used to protect unequally the data packet stream.

First, we propose two extensions to the standard Pro-MPEG COP3 codes. The first one is an equal error protection (EEP) scheme that allows the use of up to three interleaving depths to boost further channel adaptation and error decorrelation. The second one is a low-complexity unequal error protection (UEP) framework that allows to allocate unequally the available protection resources among different sets of data packets regarding their relevance. To do that, the proposed approach enables the use of not only one, as in the standard Pro-MPEG COP3 codes, but a number of matrices of dissimilar dimensions per protection block. The use of this extension allows to apply uneven code rates to unevenly important data units.

Then, we present a procedure for optimizing the selection of suitable configurations to protect data when using the proposed UEP framework. This optimization strategy is based on the hybridization of two very well known metaheuristics: simulated annealing (SA) and tabu search (TS), whose core procedures are modified to fit the characteristics of the considered scenario, and so find near-optimal solutions when strict time restrictions apply. This strategy takes as input the importance of the video data packets and the available resources, and finds in real time near-optimal configurations to protect them, in terms of number of matrices to be used and their dimensions.

Finally, we introduce a full-context-aware, yet lightweight distortion model. The proposed approach estimates the contribution of any given video packet to the resulting overall expected distortion of the sequence, and thus, its actual relevance, by taking into account upper-level characteristics of the encoded video streams and the behavior of the communication channel. This contribution is estimated considering not only the image degradation that its loss will potentially introduce in the sequence and the likelihood of this event occurring, but also the effect that the potential loss of data in reference frames/slices have on the actual importance of that packet. Download here