Page 136 - ITU Journal, ICT Discoveries, Volume 3, No. 1, June 2020 Special issue: The future of video and immersive media
P. 136

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



             Table 2 – Coding performance comparison of LCEVC
                 (LTM 4.0) over AVC/HEVC anchor (JM/HM)
                                                                    8
              Video data set    PSNR     VMAF      MOS              6
           (base & anchor codec)                                 MOS  4
            Avg. A (H.264/AVC)   −36.46%   −43.44%   −44.98%        2
           Avg. A (H.265/HEVC)   −8.77%   −24.34%   −43.05%
            Avg. B (H.264/AVC)   12.40%   −20.68%   −31.98%         0
                                                                      0         10,000       20,000      30,000
          Subjective tests using formal Mean Opinion Scores                         Bit rate (kbps)
          (MOS)  according  to  ITU-R  BT.2095  [11]  were              Fig. 4 – RD-curves for sequence A1 using
          performed by the GBTech Laboratories under the                 H.264/AVC as base and anchor codec
          supervision  of  Vittorio  Baroncini,  Chair  of  MPEG
          Test  Group.  The  results  highlight  greater  MOS
          benefits than what has been suggested by objective                     ParkRunning3
          metrics. The results per sequence are also shown in      35
          Table 2 and were calculated using the Bjøntegaard        33
          metric.  The  better  correlation  of  VMAF  with      PSNR (dB)  31
          subjective MOS results is likely due to VMAF having      29
          been  designed  to  assess  the  visual  quality  of
          encodes  at  different  resolutions  with  the  “convex   27  0       10,000       20,000      30,000
          hull” methodology [12], and thus in the presence of                       Bit rate (kbps)
          spatial scaling. PSNR is known to be less reliable at
          approximating  subjective  assessments  when             90
          comparing  different  codecs,  especially  in  the
          presence of scaling [13][14].                            80
          Figures 4 to 11 show the rate-distortion (RD) curves   VMAF  70
          for  each  sequence  in  the  data  set  A  using  the   60
          reference  implementation  of  LCEVC,  as  well  as      50
          H.264/AVC and H.265/HEVC as the base and anchor             0         10,000       20,000      30,000
                                                                                    Bit rate (kbps)
          codecs.
                      LTM 4 (JM/HM)           JM/HM                 6

                                                                    4
                             Campfire                            MOS
             38                                                     2

            PSNR (dB)  36                                           0  0        10,000       20,000      30,000

             34
                                                                                    Bit rate (kbps)
             32                                                         Fig. 5 – RD-curves for sequence A2 using
                0         10,000       20,000      30,000
                              Bit rate (kbps)                            H.264/AVC as base and anchor codec

            100                                                                  FoodMarket4
                                                                   40
             90                                                    38
            VMAF  80                                             PSNR (dB)  36
                                                                   34
             70                                                    32
                                                                   30
                0         10,000       20,000      30,000          28
                              Bit rate (kbps)                         0          5,000       10,000      15,000
                                                                                    Bit rate (kbps)







          114                                   © International Telecommunication Union, 2020
   131   132   133   134   135   136   137   138   139   140   141