ITU-R BT 2017-1998 Steroscopic Television MPEG-2 Multi-View Profile《立体电视MPEG-2多视角图像》.pdf

《ITU-R BT 2017-1998 Steroscopic Television MPEG-2 Multi-View Profile《立体电视MPEG-2多视角图像》.pdf》由会员分享，可在线阅读，更多相关《ITU-R BT 2017-1998 Steroscopic Television MPEG-2 Multi-View Profile《立体电视MPEG-2多视角图像》.pdf（7页珍藏版）》请在麦多课文档分享上搜索。

1、REPORT ITU-R BT.2017 STEREOSCOPIC TELEVISION MPEG-2 MULTI-VIEW PROFILE (1998) 1 The extension of the MPEG-2 video standard (ITU-T Recommendation H.262 i ISOICEI 13818-2: Information technol- ogy - Generic coding of moving pictures and associated audio information: Video) for multi-view applications

2、(e.g. used for stereoscopic video) has been promoted to a final International Standard at the ISOLEC JTC I/SC 29NG 11 meeting in September 1996 (Amendment 3, WG 1 i N1366) the multi-view profile (MVP) is envisioned to be a profile appropriate for applications that require multiple viewpoints within

3、the context the MPEG-2 video standard, MVP supports stereoscopic pictures as its source images for a wide range of picture resolution and quality as requested by the applications to be used. Overview of MPEG-2 multi-view profile (MVP) 1.1 Coding scheme for MVP A block diagram of the codec reference

4、model for the MVP is shown in Fig. 1. Its main features are a monoscopic coding in its base layer for compatibility and a hybrid prediction of motion and disparity for compression efficiency. Temporal scalability tools are used for coding an enhancement layer. FIGURE 1 The codec reference model for

5、the MVP Motion and disparity compensated DCT encoder Motion and disparity compensated DCT decoder I In right Disparity Disparity compensator Motion -I compensated DCT encoder i 4 Systen Mux - System Demux Motion compensated DCT decoder Disparity vectors Disparity compensator Out right - Out left c c

6、ompensated DCT decoder DCT : discret cosine transform Rap 2017-01 A configuration of prediction modes are shown in Fig. 2. A monoscopic coding with the same tools as main profile (MP), including the ISO/IEC 11 172-2 Standard is applied to the base layer. A base layer of MVP is assigned to a left vie

7、w and an enhancement layer is assigned to a right view. An enhancement layer is coded using temporal scalability tools and a hybrid prediction of motions and disparity can be utilized in the enhanced layer. It foresees higher compression of the right view of stereoscopic video by exploiting the simi

8、larity between the left and right views. MVP, one of the scalable profiles in terms of multiple viewpoint layers, has the same type of compatibility features; other scalable profiles have such compatibility with MP. For example: - decoders compliant to MVP at a certain level are capable of decoding

9、the bitstreams compliant to MP at the corresponding level (i.e. forward compatibility) decoders compliant to MP at a certain level are capable of decoding the bitstream in the base layer of MVP (Le. backward compatibility). - STD-ITU-R RECMN BT-2037-ENGL 1998 4855232 053b482 055 O 2 Rep. ITU-R BT.20

10、17 Enhancement (right view) FIGURE 2 Prediction configuration example with M = 3 coding of left-view, right-view frame picture coded using disparity prediction with respect to left-view and motion prediction with respect to itself Samples/line 1920 Linedframe I 1152 Frameds 60 Rap 2t7-02 Lower (left

11、 view) Enhancement (right view) Lower (left view) 1.2 Parameter values for MVP The levels for the MVP are high, high-1440, main and low. Temporal scalability involves two layers, a base layer and an enhancement layer. Both the enhancement and base layers have the same spatial resolution at the same

12、frame rate. Tables 1 to 4 present bounds on sampling rates, luminance pel rates, bit rates and buffer sizes for the MVP. Samples/line 1920 Linedframe 1152 Frameds 60 Sampleslline 1440 Lineslframe 1152 Framesis 60 Samples/line 1440 Linedframe 1152 Frames/s 60 TABLE 1 Upper bounds for sampling density

13、 Lower (left view) Enhancement (right view) Lower (left view) Level High Samples/line 720 Linedframe 576 Framesls 30 Samples/line 352 Linedframe 288 Frameds 30 Sampledline 352 Lineslframe 288 Frameds 30 High- 1440 Main Low Spatial resolution layer I Multiview Enhancement (right view) Samples/line 72

14、0 Lineslframe I 576 Frames/s 30 Rep. ITU-R BT.2017 TABLE 2 Upper bounds for luminance sample rate (sampleds) High High-1440 I 1 ixvei Enhancement (right view) Lower (left view) Enhancement (right view) Lower (left view) 62 668 800 62 668 800 47 O01 600 47 O01 600 Multiview Spatial resolution layer M

15、ain Low Enhancement (right view) Lower (left view) Enhancement (right view) Lower (left view) 10 368 O00 10 368 000 3 041 280 3 041 280 High High- 1440 Main LOW TABLE 3 Upper bounds for bit rates (Mbis) 130 both layers 80 base layer 100 both layers 60 base layer 25 both layers 15 base layer 8 both l

16、ayers 4 base layer Profile Multiview Level Layer Multiview TABLE 4 Buffer size requirements (bits) High High- 1440 Main Enhancement 15 898 480 Base 9 787 248 Enhancement 12 222 464 Base 7 340 032 Enhancement 3 047 424 LOW Base 1 835 008 Enhancement 950 272 Base 475 136 3 1.3 Camera parameter extensi

17、on An extension for camera information has been introduced in MVP. The extension specifies the height of image device, the focal length, the F-number, the vertical angle of the field of view, the position and the direction of the camera, and upper direction of the camera. STD-ITU-R RECMN BT-2017-ENG

18、L L998 I 4855232 053b484 928 = 4 Rep. ITU-R BT.2017 Sequences Algorithms and bit-rates (lefvnght view) . I 2 Assessment tests for MVP “Street organ”, “Flower pot”, “Trapeze” (525/60) “Fun fair” (625/50) MVPML: 613 Mbit/s, 914 Mbit/s Simulcast of MPML: 4274.5 Mbitk, 6.Y6.5 Mbit/s Simulcast of MPML as

19、 lower anchor: 2.5/2.5Mbit/s (for “Street organ”, “Fun fair”), i .5/1.5 Mbit/s (for “Flower pot”, “Trapeze”) OriginaVoriginal as upper anchor The verification tests for the MVP were carried out at three different test sites located in Japan, Germany and Canada. The results of tests were presented at

20、 the WG 11 Chicago meeting (WG 11 N1373, September 1996. Test and video subgroup “Results of MPEG-2 multiview profile verification test”). The results of the different test sites are consistent with each other and show that in general, at the tested bit rates, the observers judged that the MPEG-2 mu

21、lti-view profile coding scheme did not introduce annoying coding artifacts. Observers 2.1 Test method HHI: 24 non-expert viewers CRC: i 8 non-expert viewers NHK: 19 non-expert viewers (an observer was rejected by screening based on Recomnebdation TU-R BT.500) The double stimulus impairment scale met

22、hod (variant II) in Recommendation ITU-R BT.500 was applied. Instead of the discrete scale recommended by ITU-R a continuous scale was used in order to obtain more precise evaluations. 2.2 Test conditions The test sequences generated during bit stream exchange were used. An overview of the test cond

23、itions is provided in Table 5. Different display systems were used at each test site. TABLE 5 Overview of the subjective test conditions Test method Stereoscopic display system (picture size, viewing distance) The double-stimulus impairment scale method (variant II) described in Recommendation ITU-R

24、 BT.500, with a continuous scale HHI: two-mirror display system (19 cm x 14 cm, 5 H) CRC: time sequential display and LCD shutter eyeglasses (40.6 cm x 30.5 cm, 4 H) NHK: LCD high definition television (HDTV) projectors and polarizing eyeglasses (82 cm x 57 cm, 5 H) 2.3 Results of subjective assessm

25、ent tests The mean scores and the 95% confidence intervals of the means were calculated for each test condition. The test results of HHI, CRC and NHK are provided in Table 6 and Fig. 3. “11 and HHI2 are results obtained at HHI on two different parts of the same sequence. HI3 could not test the whole

26、 sequences because of limitations of display memory size. STD-ITU-R RECMN BT-2017-ENGL 1998 = 4855212 053b48-i b4 I Rep. ITU-R BT.2017 “12 TABLE 6 4.86 4.68 4.85 4.44 4.24 1.80 k0.13 10.19 10.13 r10.24 k0.32 k0.23 Mean scores and the 95% confidence intervals “12 4.83 4.48 4.33 4.08 4.16 1.69 k0.14 f

27、0.21 10.22 k0.26 f0.25 k 0.24 HHI 1 c) Sequence: Trapeze Source MVP MPx2 MVP MPx2 Lower (914 Mbit/s) (636.5 Mbitis) (6/3 Mbit/s) (434.5 Mbitis) anchor 4.83 3.96 4.23 3.10 3.46 1.27 10.14 f 0.32 rt 0.29 k0.35 k0.27 k 0.20 5 STD-ITU-R RECMN BT.ZDL7-ENGL 1998 4855232 0536486 7T0 m 6 Rep. ITU-R BT.2017

28、FIGURE 3 Mean scores of subjective assessment 50 5.0 4.0 4.0 3.0 3.0 2.0 2.0 1 .o 1 .O NHK CRC “11 “12 a) Street organ NHK CRC “11 “12 b) Flower pot 50 5.0 4.0 4.0 3.0 3 .O 20 2.0 1 .O 10 NHK CRC HHI c) Trapeze Source MVP (6/3 Mbits) MVP (9/4 Mbits) 0 MP x 2 (6.5i6.5 Mbit/s) MP x 2 (4.5/4.5 Mbit/s)

29、Lower anchor HHI d) Fun fair Some aspects of the results of this test seem worth being mentioned: - Within each of the four sequences, the mean score of the MVP sequence at a bit rate of 914 Mbits does not differ significantly from the mean score of the simulcast of Mps at a bit rate of 6.5/6.5 Mbit

30、s. As well, the mean score of the MVP sequence at a bit rate of 6/3 Mbit/s does not differ significantly from the mean score of the simulcast of MPs at a bit rate of 4.Y4.5 Mbitfs, except the one pair of the sequence “Street organ.” For “Street organ,” the quality of MVP is superior to that of simul

31、cast of MPs. These results show that differences in subjective evaluation between MVP and simulcast of MPs are very small at higher bit rate for pictures with slight motions (“Flower pot” and “Trapeze”) andor with significant luminance difference between left and right views (“Fun fair”). - ”Fun fai

32、r” is the scene with the most differing mean scores. In this scene the most movement (changes of the image content to the next frame) could be observed in comparison with the other scenes. Especially in “Fun fair”, the moving objects cover most of the image. 3 Future work on stereoscopic television

33、Progress made up to now has provided evidence that stereoscopic television is technically feasible. The recently approved MPEG multi-view profile provides a basis for coding and compression of stereoscopic video Sequences. The quality assessment tests carried out also have brought evidence that, wit

34、hin the limits of the test parameters chosen, STD-ITU-R RECMN BT.2017-ENGL 1998 m 4855212 D53b487 b37 m Rep. ITU-R BT.2017 7 subjectively perceived satisfactory picture quality can be achieved. Nevertheless, still many issues remain unanswered. Some of the issues that require further information are

35、 as follows: 3.1 Requirement - It will be desirable that any future stereoscopic television system is compatible with the currently emerging monoscopic digital television systems, and additional bit rate should be as small as possible. The quality of the monoscopic main picture that may be viewed on

36、 a monoscopic television display should be as close to that of the quality of a monoscopic picture using the entire channel capacity. - 3.2 - Required information for both standard definition television (SDTV) and HDTV The degree of asymmetric bit-rate allocation that is possible to the left- and ri

37、ght-view pictures for a stereoscopic video sequence to achieve minimal quality degradation for the base level picture. The effect of asymmetric bit rate allocation to the left- and right-view pictures on the subjectively perceived coding and compression artefacts and overall quality of the stereosco

38、pic video sequence. The factors that may lead to viewer fatigue; and mitigating measures that could reduce or eliminate such fatigue. The required bit rate range to achieve subjectively perceived satisfactory quality for both the stereoscopic picture as number of video sequences representing a wide

39、range of programming material and for a wide bit-rate range. Appropriate assessment test methods for stereoscopic images. Coding algorithms that enable more efficient compression of stereoscopic television signais. - - - I well as the monoscopic picture provided by the base level picture, through additional assessment tests with a large - - These studies should be carried out maintaining the liaison with WP 11B, JWP 10-llQ, and other relevant working parties and organizations.