Multiview Depth Map Enhancement by Variational Bayes Inference Estimation of Dirichlet Mixture Models
2013 (English)In: 2013 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), IEEE , 2013, 1528-1532 p.Conference paper (Refereed)
High quality view synthesis is a prerequisite for future free-viewpointtelevision. It will enable viewers to move freely in a dynamicreal world scene. Depth image based rendering algorithms willplay a pivotal role when synthesizing an arbitrary number of novelviews by using a subset of captured views and corresponding depthmaps only. Usually, each depth map is estimated individually bystereo-matching algorithms and, hence, shows lack of inter-viewconsistency. This inconsistency affects the quality of view synthesis negatively. This paper enhances the inter-view consistency ofmultiview depth imagery. First, our approach classiﬁes the colorinformation in the multiview color imagery by modeling color witha mixture of Dirichlet distributions where the model parameters areestimated in a Bayesian framework with variational inference. Second, using the resulting color clusters, we classify the correspondingdepth values in the multiview depth imagery. Each clustered depthimage is subject to further sub-clustering. Finally, the resultingmean of each sub-cluster is used to enhance the depth imagery atmultiple viewpoints. Experiments show that our approach improvesthe average quality of virtual views by up to 0.8 dB when comparedto views synthesized by using conventionally estimated depth maps.
Place, publisher, year, edition, pages
IEEE , 2013. 1528-1532 p.
, IEEE International Conference on Acoustics, Speech and Signal Processing. Proceedings, ISSN 1520-6149
Multiview video, depth map enhancement, variational Bayesian inference, Dirichlet mixture model
IdentifiersURN: urn:nbn:se:kth:diva-120338DOI: 10.1109/ICASSP.2013.6637907ISI: 000329611501142ScopusID: 2-s2.0-84890452616ISBN: 978-1-4799-0356-6OAI: oai:DiVA.org:kth-120338DiVA: diva2:614372
The 38th International Conference on Acoustics, Speech, and Signal Processing (ICASSP; Vancouver Convention & Exhibition Centre, Vancouver, Canada, on May 26 - 31, 2013.
QC 201402242013-04-042013-04-042014-02-24Bibliographically approved