Depth-based 3D video allows to synthesize additional views for auto-stereoscopic displays. For synthesized view distortion (SVD) aware compression of the depth data, prior methods consider only an SVD estimate. This thesis targets a different approach: Rendering of the synthesized view and direct computation of the exact SVD. This approach raises several questions, which this thesis addresses in two parts. The first part addresses the questions how to define a rendering-based SVD measure and how to implement it efficiently. Therefore, the first part develops the concept of the synthesized view distortion change (SVDC), which allows an additive and block related SVD measurement. The SVDC is the distortion change of the synthesized view that is caused by a depth change and can thus be determined by re-rendering the affected synthesized view texture regions. As computational resources for rendering in encoding are limited, the first part develops a low-complexity rendering algorithm by selecting, combining and optimizing basic approaches. The algorithms generates synthesized views with high quality for undistorted depth and sufficient rendering quality for distorted depth. The algorithm can be applied as receiver-side renderer. Key feature of the algorithm is that it is easily extendable by methods for partial re-rendering, which are required for SVD computation in encoding. To implement these, the first part analyzes which synthesized view regions actually change when the depth map changes and modifies the low-complexity rendering algorithm to only re-render them. Finally, the first part embeds the algorithm in a framework, called renderer model (RM), which holds the current state of the rendering process and provides an interface for depth encoders for SVDC computation. The second part addresses the question how to use SVDC computation with the RM optimally in depth encoding. An initial analysis of the encoder shows how coding efficiency and mode selection statistics change when using the SVDC instead of the depth distortion. Based on this, the second part further optimizes and analyzes complexity and coding efficiency of SVDC-based encoding while considering: 1) the impact the techniques used by the RM, 2) different options to generate the tested and the reference synthesized view textures used for SVDC computation (e.g. view extrapolation and interpolation using coded or uncoded data); 3) an additional depth distortion term; 4) the usage of the SVDC in different coding mode selection stages, and 5) the selection of Lagrange multipliers and quantization parameters. Finally, the second part optimizes and evaluates SVDC-based encoding in different application scenarios (i.e. for varying numbers of synthesized views and different receiver-side rendering approaches), using different distortion metrics, and in comparison to estimation methods. The contributions of this thesis enable an exact SVD measurement and shows how this can be exploited in encoding. The presented approach thus increases compression efficiency for encoding of depth maps and improves the transmission of 3D video for systems with auto-stereoscopic displays.


    Zugriff

    Download

    Verfügbarkeit in meiner Bibliothek prüfen


    Exportieren, teilen und zitieren



    Titel :

    Synthesized view distortion computation by rendering for depth map encoding


    Weitere Titelangaben:

    Rendering-basierte Berechnung der Verzerrung synthetisierter Ansichten in der Tiefenkarten-Codierung


    Beteiligte:
    Tech, Gerhard (Autor:in)

    Erscheinungsdatum :

    2019



    Medientyp :

    Sonstige


    Format :

    Elektronische Ressource


    Sprache :

    Englisch



    Klassifikation :

    DDC:    629



    Edge-aided virtual view rendering for multiview video plus depth

    Muddala, Suryanarayana Murthy / Sjöström, Mårten / Olsson, Roger et al. | BASE | 2013

    Freier Zugriff


    Depth distortion under calibration uncertainty

    Cheong, L. F. / Peh, C. H. | British Library Online Contents | 2004


    ENHANCED RENDERING OF SURROUND VIEW IMAGES

    MUELLER MARTIN FRITZ / DABRAL SHASHANK | Europäisches Patentamt | 2020

    Freier Zugriff

    Enhanced rendering of surround view images

    MUELLER MARTIN FRITZ / DABRAL SHASHANK | Europäisches Patentamt | 2022

    Freier Zugriff