Vincent Jantet

Contact Information

Full curriculum vitae: (HTML) - (PDF) Email: Address: IRISA Campus Universitaire de Beaulieu 35042 Rennes Cedex - FRANCE Phone: +33 2.99.84.73.60 Fax: +33 2.99.84.71.71 Project Assistant: (Huguette Béchu) +33 2.99.84.72.28 Office: F137 Bleu

Profile

2008-2011 - PhD Thesis in the TEMICS team (IRISA), on the subject of Multi-View Video Compression.

2008 - Master Degree in Computer Science at the IFSIC, Université de Rennes 1.

Full curriculum vitae: (HTML) - (PDF)

Research Activities

3DTV and Free Viewpoint Video (FVV) are two emerging features expected by end-users for next few years. Existing 3D video representations are either image-based, geometry-based, or an hybrid representation between both of them. Pure image-based representations produce high qualities rendering result, but they need a dense real-world acquisition and a large bandwidth. On the opposite, pure geometry-based representations permit some efficient compression method, but the visual quality is degraded. The multi-view plus depth representation is the most studied 3D video format proposed by the MPEG group. Associated with a view synthesis method (VSRS), it permits a high rendering quality, but also contains high data redundancies. To reduce these redundancies, we design a compact representation, based on Layered Depth Images (LDI) proposed by Shade et Al.

In this representation, pixels are no more composed by a single color and a single depth value, but can contain several colors and associated depth values. This representation reduces efficiently the multi-view video size, and offers a fast photo-realistic rendering, even with complex scene geometry. Classical procedure to generate LDI from natural multi-view images produces layers which are still correlated. Some pixels are redundant across layers. This study - supported by the Futurim@ges project - explores a innovative LDI construction scheme, based on an Incremental process (I-LDI), designed to reduce the inter-layer correlation. The number of layers is significantly reduced for an equivalent final rendering quality.

Some techniques are explored to overcome visual artifacts, like disocclusion, cracking and ghosting.

• The LDI representation itself is one solution against disocclusion. Information about occluded texture are stored in extra layers, and used during rendering stage to fill disocclusion holes.
• Many cracks can be observed in the rendered texture, due to sampling and pixelizing. The implemented solution is based on a pixel interpolation (inpainting), restricted to isolated unknown pixels. Detected cracks are filled with the median color estimated on a sliding window.
• Some ghosting effect could appear around depth discontinuity, because pixels along object boundaries receive the contribution from both foreground and background colors. An Edge detector performed on the depth map, followed by a local foreground / background classification, permit to isolate potentially blended pixels, and to use them carefully.

Tests on Breakdancers and Ballet MVD data sets show that extra layers in I-LDI contain only 10% of first layer pixels, compared to 50% for LDI. I-LDI Layers are also more compact, with a less spread pixel distribution, and thus easier to compress than LDI. Visual rendering is of similar quality with I-LDI and LDI.

In future work, we will investigate an improved disocclusion detection to insert into the I-LDI all occluded textures. An alpha merging approach will be used to reduce luminosity discontinuity artifacts. Finally, the compression stage will be investigated with a full video sequence.