Page 116 - ITU Journal, ICT Discoveries, Volume 3, No. 1, June 2020 Special issue: The future of video and immersive media
P. 116

ITU Journal: ICT Discoveries, Vol. 3(1), June 2020




          not employ any type of prediction structure, nor uses  Table 2 – Bjøntegaard delta rate (BD-BR (%)) and BD-PSNR (dB): Bit-
                                                               rate savings and average PSNR difference comparison (reference in
          depth maps, which is enough for very high-density light
                                                               boldface).
          fields. Its design relies on a very simple R-D governed
          pipeline. These results have been reported in [17].                 PSNR-YUV BD-BR (%)
                                                                                Greek  Sideboard  Tarot  Set2
          The performance of the 4DPM, which relies on the hier-
                                                                   4DPM vs x265  -33.47  -29.59  22.37  -32.34
          archical depth-based view reconstruction scheme, is il-
                                                                               PSNR-Y BD-BR (%)
          lustrated in Fig. 10. The PSNR-YUV results are reported
          for the high-density camera-array (HDCA) images Greek,                 Greek  Sideboard  Tarot  Set2
          Sideboard, Tarot, and Set2 [18]. Table 2 shows the Bjøn-  4DPM vs x265  -5.12  -17.06   68.23  -1.28
          tegaard delta rate (BD-BR (%)) and the Bjøntegaard delta
                                                                             PSNR-YUV BD-PSNR (dB)
          PSNR (BD-PSNR (dB) [20, 19] regarding the PSNR-Y and
                                                                                 Greek  Sideboard  Tarot  Set2
          PSNR-YUV results obtained for the HDCA images. The
          HDCA images correspond to light fields with wide base-   4DPM vs x265   0.80    0.91    -0.71  1.41
          lines, offering less redundancy between the views, when             PSNR-Y BD-PSNR (dB)
          compared to the lenslet-based light fields. The 4DPM out-              Greek  Sideboard  Tarot  Set2
          performs x265 especially at the low rates due to the effi-
                                                                   4DPM vs x265   0.17    0.46    -1.56  0.32
          cient depth-based view reconstruction scheme. The com-
          parisons have been conducted according to the Common
                                                               Two groups of conformance tests are currently available.
          Test Conditions specifications and validated by ISO/IEC
                                                               These groups retain a subset of features from JPEG Pleno
          SC29 WG1 experts.
                                                               Part 2 standard, which depends on the encoding mode.
          Table 1 – Bjøntegaard delta rate (BD-BR (%)) and BD-PSNR (dB): Bit-  The first group of tests corresponds to the 4DTM tool. The
          rate savings and average PSNR difference comparison (reference in  second group of tests addresses the 4DPM tool.
          boldface).
                                                               During the decoder conformance testing, a procedure is
                        PSNR-YUV BD-BR (%)
                                                               executed to evaluate the conformance according to the
                           Bikes  Danger  Fountain  Pillars
                                                               conformance class the decoder belongs to, given a spe-
             4DTM vs x265  -32.15  -36.37  -6.22  -30.24       cific encoding mode. conformance classes define the con-
             4DPM vs x265  -0.32  -13.65  -5.86   -13.07
                                                               straints to which a given JPEG Pleno implementation is
             4DTM vs 4DPM  -31.47  -21.97  -11.15  -10.51
                                                               subjected.
                          PSNR-Y BD-BR (%)
                           Bikes  Danger  Fountain  Pillars    The procedures are designed based on abstract test suites
             4DTM vs x265  -19.02  -24.18  -13.04  -20.23      (ATS) specifications, which define general tests for com-
             4DPM vs x265   0.80  -14.87  -8.59   -11.39       ponents of parts 1 and 2 of the JPEG Pleno standard. Each
             4DTM vs 4DPM  -22.37  -7.06   3.08   -0.83        ATS defines test purposes, methods, and references of the
                       PSNR-YUV BD-PSNR (dB)                   portion of the document that is being tested. The practi-
                                                               cal embodiment of the ATS are the executable test suites
                           Bikes  Danger  Fountain  Pillars
                                                               (ETS). Commonly ATS are embodied into one ETS, which
             4DTM vs x265   1.41  1.46    0.54    0.96
             4DPM vs x265   0.31  0.62    0.33    0.43         consists of codestreams, reference decoded images, a tex-
             4DTM vs 4DPM   0.95  0.69    0.26    0.31         tual description of the contents of the codestream and tol-
                                                               erance values for MSE and peak error.
                        PSNR-Y BD-PSNR (dB)
                           Bikes  Danger  Fountain  Pillars
                                                               4.4 JPEG Pleno Part 4: Reference software
             4DTM vs x265   1.06  1.11    0.11    0.73
             4DPM vs x265   0.29  0.77    0.52    0.49         For the purpose of testing JPEG Pleno encoder and de-
             4DTM vs 4DPM   0.72  0.24    -0.18   0.07
                                                               coder implementation and to help those implementing
                                                               adequately understand the algorithms and methods of
                                                               Parts 1 and 2, the JPEG Committee provides a reference
          4.3 JPEG Pleno Part 3: Conformance testing           software in JPEG Pleno Part 4 (ISO/IEC 21794-4). The
                                                               source code of this reference software is informative only.
          JPEG Committee has established JPEG Pleno Part 3 for
                                                               It is entitled Jpeg Pleno Model (JPLM) and has been devel-
          describing the conformance testing methodology used to
                                                               oped to ensure a fast proliferation of the JPEG Pleno stan-
          ensure that an application complies with the standard. It
                                                               dard and to be utilized in the conformance testing proce-
          provides a framework to enable conformity testing and
                                                               dure described in Section 4.3.
          includes, for instance, a description of expected function-
          ality, quality, and speed performance testing. Since the  The JPLM was designed with a focus on maintainabil-
          JPEG Pleno standard includes a considerable bulk of fea-  ity, integrability, and extensibility rather than encoding
          tures, which can be used for enabling diverse applica-  and decoding speeds. These non-functional requirements
          tions, a series of conformance tests are being defined.  were prioritized because the JPEG Pleno is a complex mul-




           94                                © International Telecommunication Union, 2020
   111   112   113   114   115   116   117   118   119   120   121