Back to search

University of Ioannina

Resource allocation and visual quality estimation for wireless video transmission

Abstract

dc:description

Providing the desired Quality of Service (QoS) as well as the maximum Quality of Experience(QoE) or improving the efficiency of H.264/AVC video transmissions over wirelessnetworks presents several challenges due to the characteristics of wireless networks,such as limited bandwidth, time-varying channel conditions, heterogeneous users, etc.In such networks, data are transmitted via the wireless radiomedium, which is a sharedmedium over which many users compete for resources. Due to the existence of manyusers, it is important to allocate resources in a fair manner among them. Resourceallocation is applied under various network infrastructures such as cellular networks,relay channels, ultrawideband networks etc.In the present thesis, we restrict our attention to the problem of resource allocationover wireless Visual Sensor Networks (VSNs), which consist of spatially distributedvideo cameras that are capable of compressing and transmitting the video sequencesthey acquire. Our goal is to ameliorate the video quality that reaches the end-userthrough efficient resource management. Specifically, we consider a Direct-SequenceCode Division Multiple Access (DS-CDMA) VSN, which employs a cross-layer design,where each node has its individual requirements in compression bitrate and energyconsumption, depending on the characteristics of the monitored scenes.The constraint that holds for each node of the network is that it has an availablebitrate that can be shared between source and channel coding and an available powerthat can be used for video sensing, processing, and transmission. Hence, the source coding rates, channel coding rates and power levels are the parameters that shouldbe optimally determined for each node, in an effort to tradeoff the video quality ofthe received videos and system’s efficacy. The source and channel coding rates cantake discrete values, while for the power levels we assume both the cases of takingcontinuous and discrete values.In order to optimally and jointly allocate system resources to all nodes, we consider four optimization criteria. Two of them aim at video distortion minimization, while therest seek for a distribution rule that offers fair utility allocations. The first one, calledthe Minimized Average Distortion (MAD), minimizes the overall average video distortionof the network, neglecting fairness among the nodes. The second criterion, called theMinimized Maximum Distortion (MMD), minimizes the maximum distortion among allnodes of the network, promoting a rather unbiased treatment of the nodes. Since thesimultaneous maximization of the video qualities of all nodes is not possible, we alsoapply cooperative game theory. Specifically, we use the Nash Bargaining Solution (NBS)in order to pinpoint one of the infinite Pareto-optimal solutions, based on the stipulationthat the solution should satisfy four fairness axioms. An additional solution extractedfrom the area of game theory that we utilize to the same problem of resource allocationis the Kalai-Smorodinsky Bargaining Solution (KSBS), which has also to comply withfour fairness axioms, and is applied to non-convex utility spaces. For comparisonpurposes, we also employ a criterion that Maximizes the Total system Utility (MTU)achieved by all nodes of the network.Special attention is also given to the solution methodology followed by all exploredoptimization criteria. For the case where the power levels assume continuous values,we propose the use of the Particle Swarm Optimization (PSO) algorithm, which is a computationalintelligence algorithm that draws inspiration from social dynamics. Also, weintroduce a hybrid algorithm, denoted as Hybrid Particle Swarm Optimization ActiveSet (HPSOAS), which combines PSO with Active Set (AS) and aims at exploiting thebenefits of the two aforementioned methods, thereby increasing efficiency. For comparisonreasons, the performance of the Interior Point (IP) and Trust Region Reflective(TRR) methods is also evaluated to the same optimization problem, when all of the optimizationcriteria are used, except for the KSBS. The KSBS is found directly from thegraphical representations of the utility sets, by following a geometric approach. In aneffort to evaluate the results offered by each optimization criterion, we invoke four differentfairness metrics: the first one considers both fairness and performance issues,and the second one measures the ‘‘equality’’ of a resource allocation (equal utilitiesfor the nodes). The third metric computes the total system utility, while the last onecomputes the total power consumption of the nodes.Another piece of the current study is focused on the joint problem of Group Of Pictures(GOP) length determination during the encoding process along with the allocationof the nodes’ transmission parameters, where the objective function is indicated by theMAD, NBS and MTU optimization criteria. In this case, we have to tackle a purelydiscrete optimization problem as it results from the discrete source and channel codingrates and the discrete power levels for each node of the network. In this vein, we usethe SARSA algorithm from the area of reinforcement learning.Since in video transmissions over lossy networks quality degradation is inevitable,a common practice is to ensure higher reliability of the crucial pieces of informationthrough the application of Unequal Error Protection (UEP). Such a scenario is also considered in our research, where based on the Cumulative Mean Squared Error (CMSE)we are able to prioritize the slices of the video sequences. In order to estimate theCMSE, we propose the use of the Least Absolute Shrinkage and Selection Operator(LASSO) regression method. A number of quality-relevant features are extracted fromthe H.264/AVC video sequences and are given as input to LASSO. Based on the estimatedCMSE values, we group the video slices into four priority classes, we assign adifferent channel coding rate to each of them, and simulate a video transmission scenarioover a noisy environment so as to investigate the performance of our proposedapproach.In the last part of this thesis, we deal with the problem of perceptual video qualityassessment. Particularly, we propose Reduced-Reference (RR) and No-Reference(NR) models so as estimate the quality of H.264/AVC video sequences, in terms of theMean Opinion Score (MOS). A variety of perceptually-motivated features are examinedto account for the impact of coding artifacts, packet losses, and video content characteristics.These features are employed for estimating video quality using the LASSOregression technique, which utilizes a subset of the input features, by selecting onlythose that have relatively higher impact on the process of video quality estimation.For comparison purposes, the Ordinary Least Squares (OLS) and the Ridge regressionmethod combined with sequential Forward Feature Selection (FFS) are also applied. In addition, performance measures as recommended by Video Quality Experts Group(VQEG) are used in order to gauge the effectiveness of our proposed models.In order to estimate subjective video quality, we usually deal with a large numberof features and a small sample set. Applying regression on complex datasets may leadto imprecise solutions due to possibly irrelevant or noisy features as well as the effectof overfitting. For this purpose, our research is extended to include a robust NR modelthat has a good generalization capability to unseen data, for videos that are impairedby both compression artifacts and packet losses. This model is able to improve theper-sequence MOS estimation accuracy, by following a frame-level MOS estimationapproach, where the MOS estimate of a sequence is obtained by averaging the perframeMOS estimates, instead of performing regression directly at the sequence level.Since it is impractical to obtain the per-frame MOS ground truth through subjectiveexperiments, we propose an objective metric able to do this task, which provides areliable indicator for the quality of each frame of a video, offering an intuition about itsindividual contribution to the overall video quality score.

Degree

thesis:*
Grantor dc:publisher
University of Ioannina
Year dc:date
2015

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Pandremmenou, Aikaterini
  • Πανδρεμμένου, Αικατερίνη

Subjects

dc:subject × 16

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/39652

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Pandremmenou, Aikaterini; Πανδρεμμένου, Αικατερίνη. Resource allocation and visual quality estimation for wireless video transmission. University of Ioannina, 2015. http://hdl.handle.net/10442/hedi/39652