1、 Rec. ITU-R BT.1663 1 RECOMMENDATION ITU-R BT.1663 Expert viewing methods to assess the quality of systems for the digital display of large screen digital imagery1in theatres (Question ITU-R 15/6) (2003) The ITU Radiocommunication Assembly, considering a) that ITU, as well as other international sta
2、ndardization bodies, is studying a new service called large screen digital imagery (LSDI); b) that several applications will likely be identified for the LSDI service; c) that it will be necessary to specify performance requirements and to check the suitability of technical solutions considered for
3、each application with the performance requirements of that application; d) that such checks will also necessarily involve subjective assessment tests under rigorous scientific conditions; e) that different subjective test methodologies may offer different sensitivities for detecting certain kinds of
4、 impairments; f) that the subjective assessment methods specified in Recommendation ITU-R BT.500 based on non-expert viewing are time-consuming and expensive, in view of the large number of scores that must be collected in order that systems may be differentiated in performance; g) that a new method
5、 of subjective testing, based on the use of a small number of expert viewers, is now proposed that will provide comparable ability to differentiate the performance of different systems while allowing faster, less expensive procedures, recommends 1 that the test method described in Annex 1, based on
6、the use of expert viewers, should be used in the subjective assessment of LSDI solutions whenever time or budget constraints do not allow use of the non-expert viewing methods specified in Recommendation ITU-R BT.500, and where the sensitivity of the method in Annex 1 is sufficient to differentiate
7、the systems being evaluated, 1Large screen digital imagery (LSDI) is a family of digital imagery systems applicable to programmes such as dramas, plays, sporting events, concerts, cultural events, etc., from capture to large screen presentation in high-resolution quality in appropriately equipped ci
8、nema theatres, halls and other venues. 2 Rec. ITU-R BT.1663 further recommends 1 that, as part of the testing process, studies be conducted to verify the sensitivity of the test method described in Annex 1; 2 that, in order to improve this Recommendation, further studies be conducted on the processi
9、ng of the results; 3 test managers and organizations are encouraged to make available to ITU administrations and Sector Members any test materials and test tools (e.g. computer programs to generate side-by-side or mirror image presentations) that are developed, in order to facilitate future testing
10、by other organizations. Annex 1 Expert viewing to assess the quality of systems for the digital display of LSDI pictures in theatres 1 Introduction In past years, expert viewing often has been employed to perform a quick verification of the performance of a generic video process. This Annex describe
11、s an expert viewing test method that will ensure consistency of results obtained in different laboratories, using a limited number of expert assessors. 2 Why a new method based on “expert viewing”? It is useful to point out the advantages resulting from the application of the proposed methodology. F
12、irst, a formal subjective assessment test typically requires use of at least 15 observers, selected as “non-experts”, requiring lengthy tests and a continuous search for new observers. This number of observers is necessary to achieve the sensitivity necessary so that the systems being tested may be
13、confidently differentiated and ranked, or be confidently judged equivalent. Second, by using non-expert observers, traditional tests may fail to reveal differences that, with protracted exposure, may become salient, even to non-experts. Third, traditional assessments typically establish measures of
14、quality (or differences in quality), but do not directly identify the artefacts or other physical manifestations that give rise to these measures. The methodology proposed here tries to solve all three problems. Rec. ITU-R BT.1663 3 3 Definition of expert subjects For the purpose of this Recommendat
15、ion, an “expert viewer” is a person that knows the material used to perform the assessment, knows “what to look at” and may or may not be deeply informed on the details of the algorithm used to process the video material to be assessed. In any case, an “expert viewer” is a person with a long experie
16、nce in the area of quality investigation, professionally engaged in the specific area addressed by the test. As an example, when organizing an “expert viewing” test session on LSDI material, experts in the production or post-production of film or in the production of high-quality video content shoul
17、d be selected (e.g. directors of photography, colour correctors, etc.); this selection has to be made considering the ability to make unique subjective judgements of LSDI image quality and compression artefacts. 4 Selection of the assessors An expert viewing test is an assessment session based on th
18、e opinions of assessors, in which judgements are provided on visual quality and/or impairment visibility. The basic group of experts is made of five to six subjects. This small number makes it easier to collect assessors, and to reach a faster decision. According to the experiment needs, it is allow
19、ed to use more than one basic group of experts, grouped into a larger combined pool of experts (e.g. coming from different laboratories). It is recognized that experts may tend to bias their scores when they test their own technology, therefore it should be avoided to include persons that were direc
20、tly involved in the development of the system(s) under test. All assessors should be screened for normal or corrected-to-normal visual acuity (Snellen Test) and normal colour vision (Ishihara Test). 5 Test material Test materials should be selected to sample the range of production values and levels
21、 of difficulty foreseen in the real context in which the system(s) under test would be used. Selection should favour more challenging material without being unduly extreme. Ideally, 5-7 test sequences should be used. The method to select material may vary also in relation to the application for whic
22、h the system under test has been designed. In this regard, no further indication is given here on rules for the selection of the test material, leaving the decision to the test designer in relation of the considerations above. 6 Viewing conditions The viewing conditions, which shall be described ful
23、ly in the test report, shall be in accordance with Table 1 and shall be kept constant during the test. 4 Rec. ITU-R BT.1663 TABLE 1 7 Methodology 7.1 Evaluation sessions Each evaluation session (defined as the set of test sittings for a given group of observers) should consist of two phases (i.e. Ph
24、ase I and Phase II). 7.1.1 Phase I Phase I consists of a formal subjective test performed in a controlled environment (see 6) which will permit valid, sensitive and repeatable test results. Here, the experts individually rate the material shown using the rating scale described below. Members of the
25、panel are not permitted to discuss what they are seeing or to control the presentations. During this phase, the experts should NOT be aware of the coding scheme under test, or of the order of presentation of the material under test. The material under test will be randomized, so as to avoid any bias
26、 in the assessment. 7.1.1.1 Presentation of material The presentation method combines elements of the simultaneous double stimulus for continuous evaluation (SDSCE) method (Recommendation ITU-R BT.500, 6.4) and the double stimulus continuous quality scale (DSCQS) method (Recommendation ITU-R BT.500,
27、 5). For reference, it may be called the simultaneous double stimulus (SDS) method. As with the SDSCE method, each trial will involve a split-screen presentation of material from two images. In most cases, one of the image sources will be the reference (i.e. source image), while the other is the tes
28、t image; in other cases, both images will be drawn from the reference image. The reference shall be the source material presented transparently (i.e. not subjected to compression other than that implicit to the source recording medium). The test material shall be the source material processed throug
29、h one of the systems under test. The bit-rate and/or quality level shall be as specified by the test design. Unlike the SDSCE method, observers will be unaware of the conditions represented by the two members of the image pair. Setting(s) Viewing conditions Minimum Maximum Screen size (m) 6 16 Viewi
30、ng distance (1)1.5 H 2 H Projector luminance (centre screen, peak white) 10 ftL 14 ftL Screen luminance (projector off) 1/1 000 of projector luminance (1)The “butterfly” presentation should be used when the viewing distance is closer to 1.5 H. If the “side-by-side” presentation is used, the viewing
31、distance should be closer to 2 H value. Rec. ITU-R BT.1663 5 The split-screen presentation shall be done either using the traditional split screen without mirroring or by the butterfly technique, where the image on the right side of the screen is flipped horizontally. Because full-width images will
32、be used, only half of each image can be displayed at a time. In each presentation, the same half of the image will be shown on each side of the display. As with the DSCQS method, the image pair is presented twice in succession, once to allow familiarization and scrutiny and once to allow confirmatio
33、n and rating. Each sequence will be 15-30 s in duration. Each sequence may be labelled at the beginning of each clip to assist assessors (see non-mirrored split screen example shown in Fig. 1). 1663-01Vote NBA1 s 1 s e.g. 20 s 1 s 1 s e.g. 20 s 4 sVoting timeFIGURE 17.1.1.2 Judgement scale The crite
34、rion for acceptability in LSDI applications is that the test (i.e. compressed) image be indistinguishable from the reference. Several commonly used scoring methods can be used to evaluate the systems under test. A suggested method is the stimulus comparison scales recommended in Recommendation ITU-R
35、 BT.500, 6.2. A specific example scale is the non-categorical (continuous) SAME-DIFFERENT scale as described in Recommendation ITU-R BT.500, 6.2.4.2: 1663-0250 cmSame DifferentFIGURE 27.1.1.3 Judgement session The session, which may involve more than one sitting depending on the number of test condi
36、tions, shall involve two types of trials: test trials and check trials. In a test trial, one half of the display shows the reference while the other half shows the test. In a check trial, both halves show the reference. The purpose of the check trial is to assess a measure of judgement bias. 6 Rec.
37、ITU-R BT.1663 For each system tested, the following test trials are required for each test sequence: TABLE 2 Preferably, there would be at least two repetitions of each of the cases above. For each system, the following check trials are required for each test sequence: TABLE 3 Again, preferably ther
38、e would be at least two repetitions of each of the cases above. The test session should be divided into sittings not more than one hour in duration separated by 15 min rest periods. Test and check trials resulting from the combination of codec and test sequence should be distributed across sittings
39、by pseudorandom assignment. It is more complex, but worthwhile, to impose some restriction on this process. For example, if there were four sittings, one might randomly assign each of the four test trials for a given codec and test sequence to a randomly determined position in one of the sittings. T
40、his approach has the benefit of ensuring that each systems test trials are distributed over the entire test session. 7.1.1.4 Processing of test scores For a given test trial, the test score is the distance between the “SAME” endpoint of the scale and the mark made by the observer, expressed on a 0-1
41、00 scale. The results will be analysed in terms of mean opinion score (MOS), and the MOS will be used to establish rank ordering of the systems tested. Depending on the number of observations per system (observers test sequences repetitions), the data may be subjected to analysis of variance (ANOVA)
42、2. Performance on check trials can be used to derive a baseline “chance” judgement difference. 2A total of 10-20 observations in the lowest-order condition of interest is sufficient for application of inferential statistical treatments, such as ANOVA. Left display panel Right display panel Left half
43、 reference Left half test Right half reference Right half test Left half test Left half reference Right half test Right half reference Left panel Right panel Left half reference Left half reference Right half reference Right half reference Rec. ITU-R BT.1663 7 7.1.2 Phase II One of the main goals of
44、 Phase II is to refine the relative ranking of the results of Phase I, the precision and reliability of which may be reduced by a limited number of observers and/or judgement trials. A further, and important, objective is to elicit observations as to the characteristics upon which images are perceiv
45、ed to differ and upon which judgements in Phase I were based. This part involves review by the expert panel of the material shown. Here, the experts are permitted to discuss the material as it is shown, to repeat part or all of the material as many times as necessary for review and/or demonstration,
46、 and to arrive at a consensus judgement and a description of what they see. “Trick Play”, including the use of modes such as slow motion, single step and still frame, are permitted if requested by the expert viewers. These techniques will require some interaction with, and intervention by, the test
47、manager. 7.1.2.1 Grouping the material under test To properly perform the Phase II test, it is necessary to group the material under test by content, obtaining a so-called Basic Expert viewing Set (BES), i.e. all the coded sequences obtained from the same source sequence have to be grouped and then
48、ordered in accordance with the ranking derived from Phase I. The test material will be ordered from the lowest MOS value to the highest MOS value. There will be as many BESs as the number of sequences used for the test. 7.1.2.2 Basic expert viewing test sub-session A basic expert viewing (BEV) test
49、sub-session is a discussion session during which the experts examine all the material included in a BES; one task is to confirm or modify the ranking order that resulted from the Phase I formal test. Therefore, the relative visibility of differences has to be confirmed or modified. 7.1.2.3 Phase II plan During Phase II, all the BEVs have to be carried out. The experts will be made aware that the presentation order is the result of the ranking of Phase I. The experts will not be aware of any relationship between proponent systems and ranking. Phase II will be conducted as a grou
copyright@ 2008-2019 麦多课文库（www.mydoc123.com）网站版权所有
备案/许可证编号：苏ICP备17064731号-1