Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.633212
Title: Gaze location prediction and enhanced error resilience
Author: Cheng, Qin
ISNI:       0000 0004 5365 1042
Awarding Body: University of Bristol
Current Institution: University of Bristol
Date of Award: 2014
Availability of Full Text:
Access through EThOS:
Abstract:
The sensitivity of the human visual system decreases dramatically with increasing distance from the fixation location in a video frame. Accurate prediction of a viewer's gaze location has the potential to improve bit allocation, rate control, error resilience and quality evaluation in video compression. Commercially, delivery of football video content is of great interest due to the very high number of consumers. In this thesis we propose a gaze location prediction system for high definition broadcast football video. The proposed system uses knowledge about the context, extracted through analysis of a gaze tracking study that we performed, in order to build a suitable prior map. We further classify the complex context into different categories through shot classification thus allowing our model to pre-learn the task pertinence of each object category and build the prior map automatically. We thus avoid the limitation of assigning the viewers a specific task, allowing our gaze prediction system to work under free-viewing conditions. Bayesian integration of bottom-up features and top-down priors is finally applied to predict the gaze locations. Results show that the prediction performance of the proposed model is better than that of other top-down models which we adapted to this context. The next part of this thesis focuses on enhancement of error resilience in the video transmission chain. Video transmission over error prone channels can suffer from packet losses when channel conditions are not favourable. As a result the distortion/quality of the decoded video at the receiver often differs from that of the encoded video at the transmitter. Accurate estimation of the end to end distortion (the distortion due to compression and packet loss after decoder error concealment) at the encoder/transmitter can lead to more efficient and effective application of error resilience (e.g. selective intra coding, forward error correction, etc.). This proposed end to end distortion estimation model incorporates a probabilistic estimation of the distortion introduced by advanced error concealment methods, which often used by decoders to mitigate t.he effect of packets loss. The proposed model offers significant improvements in estimation accuracy relative to existing models that only consider previous frame copy as the concealment strategy of the decoder. The final goal is to foveat the end to end distortion model using the predicted gaze locations to provide the optimal subjective quality of the decoded video.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.633212  DOI: Not available
Share: