Temporal Consistent Depth Map Upscaling for 3DTV
2014 (English)In: Proceedings of SPIE - The International Society for Optical Engineering: Three-Dimensional Image Processing, Measurement (3DIPM), and Applications 2014 / [ed] Atilla M. Baskurt; Robert Sitnik, 2014, Art. no. 901302- p.Conference paper (Refereed)
Precise scene depth information is a pre-requisite in three-dimen-sional television (3DTV), e.g. for high quality view synthesis in autostereoscopic multiview displays. Unfortunately, this information is not easily obtained and often of limited quality. Dedicated range sensors, such as time-of-flight (ToF) cameras, can deliver reliable depth information where (stereo-)matching fails. Nonetheless, since these sensors provide only restricted spatial resolution, sophisticated upscaling methods are sought-after, to match depth information to corresponding texture frames.Where traditional upscaling fails, novel approaches have been proposed, utilizing additional information from the texture for the depth upscaling process. We recently proposed the Edge Weighted Optimization Concept (EWOC) for ToF upscaling, using texture edges for accurate depth boundaries. In this paper we propose an important update to EWOC, dividing it into smaller incremental upscaling steps. We predict two major improvements from this. Firstly, processing time should be decreased by dividing one big calculation into several smaller steps. Secondly, we assume an increase in quality for the upscaled depth map, due to a more coherent edge detection on the video frame.In our evaluations we can show the desired effect on processing time, cutting down the calculation time more than in half. We can also show an increase in visual quality, based on objective quality metrics, compared to the original implementation as well as competing proposals.
Place, publisher, year, edition, pages
2014. Art. no. 901302- p.
, Conference volume, 9013
3d video, 3DTV, depth aqcuisition, optical flow, depth map upscaling
IdentifiersURN: urn:nbn:se:miun:diva-21927DOI: 10.1117/12.2032697ISI: 000336030900001ScopusID: 2-s2.0-84901748474OAI: oai:DiVA.org:miun-21927DiVA: diva2:716786
Three-Dimensional Image Processing, Measurement (3DIPM), and Applications 2014; San Francisco, CA; United States; 5 February 2014 through 5 February 2014; Code 105376