WO2008139135A1

WO2008139135A1 - Video signal analysis

Info

Publication number: WO2008139135A1
Application number: PCT/GB2008/001276
Authority: WO
Inventors: Andrew Gordon Davis
Original assignee: British Telecommunications Public Limited Company
Priority date: 2007-05-09
Filing date: 2008-04-11
Publication date: 2008-11-20
Also published as: KR101452541B1; JP5074580B2; EP2145469B1; EP1990991A1; JP2010526502A; US9800824B2; CN101690177B; US20100141835A1; CN101690177A; EP2145469A1; KR20100017310A

Abstract

In order to detect interlace errors in a video signal, one receives successive digitally coded frames, each frame comprising data for a field of a first type and data for a field of a second type (i.e. a top field and bottom field or vice versa). One then generates for each field of the first type: - a first difference signal (Dl F1 ) representative of the difference between the field and the second-type field of the previous frame; - a second difference signal (Dl F2) representative of the difference between the field and the second-type field of the same frame; and - a third difference signal (Dl F3) representative of the difference between the field and the second-type field of the following frame. Then, in dependence of the values of said difference signals, a decision signal (wOFIag) is generated indicating an estimated temporal relationship of the field to the second-type field of the same frame. This can be compared with a signal (IDFO) indicating a purported temporal relationship, and an alarm signal (wEFIag) generated in the event of a mismatch.

Description

Video Signal Analysis

1. Introduction

The present invention relates to the analysis of video sequences for display on interlaced systems such as those according to BT.601 [1]. In such systems, the use of incorrect field ordering (top- field-first (TFF) or bottom-field-first (BFF) can result in noticeable and objectionable effects in the display of both interlaced and progressive-type sequences.

According to the present invention there is provided a method of detecting field order of a video signal, comprising: receiving successive digitally coded frames, each frame comprising data for a field of a first type and data for a field of a second type; generating for each field of the first type:

- a first difference signal (DIF1) representative of the difference between the field and the second-type field of the previous frame; - a second difference signal (Dl F2) representative of the difference between the field and the second-type field of the same frame;

- a third difference signal (Dl F3) representative of the difference between the field and the second-type field of the following frame; and

- in dependence of the values of said difference signals, a decision signal (wOFIag) indicating an estimated temporal relationship of the field to the second-type field of the same frame.

Other aspects of the invention are set out in the claims.

Some embodiments of the invention will now be described, by way of example, with reference to the accompanying drawings, in which: Figures 1 to 10 and 12 are timing diagrams for use in explaining embodiments of the invention;

Figure 11 is a functional block diagram showing the operation of one embodiment of the invention; and

Figures 12 to 27 are graphs showing results obtained. .Here, a method is presented to detect field order problems in video content by the sliding- window comparison of a target field (top or bottom) with three neighbouring opposite- phase fields. For each analysed field, a number of measures are generated and used to make an "instant" classification of the frame containing the target field. These frame results are buffered and corresponding "windowed" decisions made. The "instant" and "windowed" results are used in a "combined" analysis technique to identify video field- ordering properties and the nature and location of potential field-ordering problems. Furthermore, the technique is enhanced by an "instant" -only method for identifying problem areas of frame-repeated progressive content. 2. Picture Structures

A brief description of interlaced and progressive video properties is given in this section prior to the description of the analysis technique in Section 3.

2.1 Interlaced Video A top-field-first interlaced video sequence may be represented as a sequence of alternating top and bottom fields as shown in Figure 1.

Temporally, the B fields should be half-way between the neighbouring T fields. Spatially, a B field will be offset vertically down by 1 line on the output display device and interlaced with the corresponding T field, as shown in Figure 2. For PAL-625 system "I" standard-resolution display (720x576 active samples per frame), a visible picture would consist of TFF interlaced 288-line T and B fields with a field update rate of 50Hz (frame-rate of 25Hz). For NTSC-525 system "M" standard resolution display (720x480 active samples per frame), a visible picture would consist of BFF interlaced 240- line T and B fields with a field update rate of 59.94 Hz (frame-rate of 29.97 Hz).

2.2 Progressive Video

Progressive video sequences have frames made up of lines from just one time-interval, such as generated by cine-film cameras and more recently digital video (DV) cameras. Considering fields presented for interlaced display, progressive sequences would have the temporal characteristics represented in Figure3, where corresponding top and bottom fields cover identical time intervals. These fields would, however, still be spatially interlaced according to Figure 2 and, in the display, temporally interlaced according to Figure 1.

Frame-repeated progressive video, as illustrated in Figure 4, might be found in cartoon content. The pattern of repeated frames is typically regular however irregular repeats are also possible.

2.3 Field Ordering Problems

There follows a brief consideration of how field-ordering properties of both interlaced and progressive content may cause problems for interlaced display.

2.3.1 Interlaced Content

The interlaced display of content with incorrect field-ordering can cause noticeable degradations. If BFF interlaced content is displayed TFF then a "juddering" effect may be visible within areas of motion. This problem is illustrated in Figures 5 and 6. 5 shows BFF field-timing for an 'X' pattern moving at a constant rate across the display.

Figure 6 shows the field-ordering for correct BFF display and also the field ordering for incorrect TFF display. The discontinuity in the motion of the 'X' is evident for the TFF display.

2.3.2 Progressive Content

One might expect there not to be a problem with field-ordering for progressive content, as a pair of fields derived for interlaced playback from one progressive frame would be from the same time interval. No motion "juddering" should occur for either field-ordering in playback. However, there are a number of video editing and processing operations that can make the interlaced playback of progressive content sensitive to field ordering.

Frame-rate conversion between cine-film (24 frames/s), PAL (25 frames/s) and NTSC (29.97 frames/s) can involve the addition and removal of fields from the original sequence. The insertion or deletion of a field within a frame causes subsequent frames to consist of fields from different time intervals, making interlaced display field-order sensitive. Field- level editing, such as cutting and fading, can also introduce such field-order sensitivity when applied to progressive content. Figure 7 shows the frame structure of progressive content with field-misalignment. Here the progressive content has been separated into fields, with matching progressive fields put in separate frames. This separation requires TFF interlaced playback and incorrect BFF playback could cause visible motion juddering. This problem will be called TFF field- misalignment, as TFF interlaced playback is required to avoid problems.

Figure 8 illustrates progressive frame-repeated content with BFF field-misalignment. In this case, TFF interlaced playback could base motion juddering.

Another example of field-misaligned progressive content, as observed in real cartoon content, is illustrated in Figure 9. Here, cartoon content, that is predominantly frame- repeated and progressive in nature, has BFF field-misalignment in frame 4 as a result of there being unequal numbers of matching top and bottom fields within the content. TFF interlaced display of such content would give motion discontinuity at this transition, which, if frequently occurring, could be perceptible.

Apart from the more persistent motion juddering, such field-level transitions can also cause visible problems at scene cuts. For the example shown in Figure 9, if the fields T1/B1 and T2/B2 belonged to scene 1 (grey) and T3/B3 onwards to scene 2 (white), then TFF and BFF display of the sequence would give the field sequences shown in Figure 10. TFF interlaced playback would give a scene change discontinuity, with a field from scene 2 played between two fields from scene 1.

3. Field-Order Analysis Algorithm

There follows a description of a technique for the detection of potentially visible field-order problems within video content subjected to interlaced display. The technique operates on either a sequence of fields or frames in a pixel intensity representation and relies on a sliding comparison between fields from 3 successive frames. Figure 11 shows a representation of the proposed process, which accepts video content frames and outputs a field-ordering error flag, wEFIag, a content decision flag, wOFIag and an instant decision flag, iOFIag.

The technique consists of two main processing paths handling "instant" (highlighted grey) and "windowed" parameter sets. The apparatus comprises two buffers Dif_buf and IDFIag_buf and ten function units Fund , Func2, etc. These could if desired be constructed as separate program-controlled processors, or if preferred the functions of two or more units could be performed by a single processor. A description of the function of each unit will now follow.

3.1 Fund -Field Difference Measure Each input video frame, video(n), may be considered to consist of an interlaced pair of top and bottom fields, T(n) and B(n). For each frame, three difference calculations are performed to indicate the match between the target field of one type, which is either the top or bottom field of the current frame, and the three second type, or opposite phase, fields of the previous, current and next frames. This process is illustrated for a bottom field target in Figure 12.

Using mean of absolute difference between pixel intensity values, the difference calculations may be defined as:

Y-I X-I

D/Fl(n) = (l/ Xy)∑ ∑abs(B(n,x, y) -nn -l,x, y)) (3.1) y=0 x=0

Y-I X-I

DlF2{ή) = (\IXY)∑ ∑abs{B{n,x, y) -T(n,x, y)) (3.2) y=0

Y-I X-I DIF3(n) = (UXY)∑ ∑abs(B(n,x, y) -T(n + l,x, y)) (3.3) y=0

In (3.1) to (3.3), B(n,x,y) represents the pixel intensity value of pixel position x horizontal and y vertical in the bottom field of the n'th frame and X and Y are the total number of horizontal and vertical pixels respectively. T(n,x,y) represents the same pixel position in the corresponding top field.

It will be noted that the difference signals represented by Equations 3.1 , 3.2 and 3.3 are differences between two fields which necessarily have a vertical spatial offset from one another which will therefore give rise to non-zero difference values even in the case where there is no motion. We have not found this to be a serious problem in practice, but reliability can be improved by applying one of the known techniques for alleviating this problem. Techniques to reject interlaced spatial artefacts, through methods such as "coring" and "diamond motion analysis" [2..9], would be of particular benefit in the matching of fields. In another approach, one can disregard individual difference values (i.e. before summation) that fall below a threshold, and/or evaluate the difference between a pixel and the corresponding pixel on both the line above it and the line below it on the other field, taking the smaller of the two.

The results quoted below used both these expedients, Equation 3.1 being replaced by

DIFl(n) = (1 / AT)∑ ∑ if min[abs(B(n, x, y) - T(n - 1, x, y)), abs(B(n, x, y) - T(n - l,x, y + 1)) y=0 Jt=O

> threshold then πάn[abs(B(n, x, y) - T(n - 1, x, y)), abs(B(n, x, y) - T(n - 1, JC, y + 1))] else 0

and similarly for Equations 3.2 and 3.3: y-i x-i

DIF2(ή) = (1/ XY j∑ ∑ if min [abs(B(n, x, y) - T{n, x, y)), abs(B(n, x, y) - T(n, x, y + 1))] > threshold then mm[abs(B(n, x, y) - T(n, x, y)), abs(B(n, x, y) -T(n,x, y + 1))] else 0

K-I X-I

DIF3(n) = (1 / XY)∑ ]T if πάn[abs(B(n, x, y) - T(n + 1, x, y)), abs(B(n, x, y) - T(n + l,x, y + 1))] . > threshold then mm[abs(B(n, x, y) - T(n + 1, x, y)), abs(B(n, x, y) - T(n + l,x, y + 1))] else 0

These compensated equations were used in the tests reported below If a top field is chosen as target then the equations are the same except that 'T' and "B" are transposed.

In this specification, references to "previous", "next", "following" or the like frames refer to the frames considered in capture-and-display order, even though a particular coding scheme may vary the order of transmission.

3.2 Func2 -Windowed Field Difference Measure

The set of 3 difference values, DIF1..3(n), are then entered into the difference buffer, where the previous K-1 sets of difference values are also stored. Func2 then performs time-averaging of the difference parameters according to equations (3.4) to (3.6).

wDlF\(ή) = (XI K)Y₄ DIFl(n - i) (3.4)

1=0 K-I wDIF2(n) = (1/ KJ∑ DIF2(n - i) (3.5) K-I wDIF3{ή) = (1/ K)∑ DIF3(n - i) (3.6)

1=0

3.3 Func3 -Windowed Decision Parameters

For each frame, the averaged difference values wDIF1..3 are ranked in order of magnitude wRanki ..wRank3, such that wRanki is equal to the maximum value and wRank3 the minimum. The variables wlndexi ..wlndex3 are set to the corresponding index numbers, so that if wDIF2 was the maximum and wDIF1 the minimum, then wlndexi would equal 2, wlndex2 would equal 3 and wlndex3 would equal 1. Windowed decision parameters are then calculated according to equations (3.7) to (3.9) wMotionFlag{ή) - {{wRank\{ή) - wRank3(n) > MThreshϊ) AND(wRankl(n) > MThreshl))

wParam\{n) = 100.0 * (wRank2(n) - wRank3(n)) l(wRank\(n) - wRank3(n)) (3.8)

wParam2(n) = 100.0 * wRank3(n) I wRank2(n) (3.9)

Possible settings for MThreshl and MThresh2 are given in Table 3.15.

The motion parameter, wMotionFlag, indicates that there is sufficient motion for a meaningful field-ordering decision to be made. This parameter depends on comparisons of absolute and relative DIF values with appropriate thresholds according to equation (3.7). Parameter, wParami , represents the difference between the 2^nd and 3^rd ranked DIF values as a percentage of the range of the DIF values and is intended to reflect the feature of two DIF values being nearly the same and significantly less than the remaining value. This parameter is based purely on differences and would be unaffected by an offset applied equally to all 3 difference values. Parameter wParam2 represents the ratio of the minimum to the middle DIF value as a percentage and is intended to reflect the overall significance of the difference between the two smallest DIF values.

3.4 Func4 -Windowed Decision Intermediate Flags ,

A decision on interlaced field-order properties for the n'th frame may be made by testing WParami and WParam2 against thresholds according to equations (3.10) and (3.11 ). wIFlagl(n) = ((wParaml(ή) < iThreshl)AND(wParam2(n) > iThresh2)

(3.10) AND(wMotionFlag («)))

wIFlag2(n) = wlndexl (3.11)

Suitable threshold values are given in Table 3.15 and the resulting flags may be interpreted according to Table 3.1.

Table 3.1 Explanation of interlaced decisions.

A decision on progressive field-order properties for the n'th frame may be made by testing wParami and wParam2 against thresholds according to equations (3.12) and (3.13). wPFlag\(ή) = {(wParaml{ή) > pThreshϊ)AND

(3.12)

(wParam2(n) < pThresh2)AND(wMotionFlag(n))

wPFlag2(n) = wlndexi (3.13)

Decision threshold pThreshi is set according to Table 3.2 and defines acceptance bands for the wParami parameter. Suitable threshold values are given in Table 3.15.

Table 3.2 Progressive decision thresholds The resulting flags may be interpreted according to Table 3.3.

Table 3.3 Explanation of progressive decisions.

The thresholds set for the interlaced and progressive decisions ensure that I and P 'True" decisions are mutually exclusive.

3.5 Func5 -Windowed Decision Flag

The intermediate decisions flags are then combined according to Table 3.4 to give a single "windowed" decision flag, wDFIag. Note that here, and in other Tables that serve to define the algorithm, the resultant of the operation is shown in the right-hand column. Tables that are merely explanatory are labelled "Explanation of ...".

Table 3.4 Combined interlaced/progressive decision flag 3.6 Func3a -Instant Parameters (iParami ,iParam2)

Func3a uses the ranking process described in Section 3.3 to produce "instant" ranked parameters, iRanki ..3 and ilndexi ..3, from difference values, DIF1(n)..DIF3(n). The resulting "instant" parameter values, iParami and iParam2, are calculated according to equations (3.14) and (3.16). iMotionFlag{n) = {(iRank\{ή) - iRank3(n) > MThreshl)

(3.14) AND(iRankl(n) > MThreshT))

iParam\{ή) = 100.0 * (iRank2(n) - iRank3(n)) /(iRankl(n) - iRank3{n)) (3.15)

iParaml{n) = 100.0 * iRank3{n) I iRank2{n) (3.16)

3.7 Func4a - "Instant" Decision Intermediate Flags

A decision on interlaced field-order properties for frame n may be made by testing iParami and iParam2 against thresholds according to equations (3.17) and (3.18). ilFlaglin) = ((iParaml(n) < iThreshϊ)AND(iParam2(n) > iThreshl)

(3.17) AND{iMotionFlag{ή)))

UFlag2(ή) = ilndexl (3.18)

The resulting flags may be interpreted according to Table 3.5.

Table 3.5 Explanation of "Instant" interlaced decisions.

A decision on progressive field-order properties for frame n may be made by testing iParami and iParam2 against thresholds according to equations (3.19) and (3.20). iPFlagl(n) = ((iParaml(n) > pThresh\)AND

(3.19)

(iParam2(n) < pThresh2)AND(iMotionFlag(n))

iPFlagl{n) = ilndexi (3.20)

Decision threshold pThreshi is set according to Table 3.2 and defines acceptance bands for the iParami parameter. The resulting flags may be interpreted according to Table 3.6.

Table 3.6 Explanation of "Instant" progressive decisions.

The thresholds set for the interlaced and progressive decisions ensure that I and P 'True" decisions are mutually exclusive and example values are given in Table 3.15 below.

3.8 Funcδa -Instant Decision Flag (iDFIaq) The "instant" intermediate decisions flags are combined according to Table 3.7 to give a single "instant" decision flag, iDFIag.

Table 3.7 Combined interlaced/progressive decision flag The instant decision flag for the n'th frame, iDFIag(n), is entered into the "Instant Flag Buffer" iDFIag_buf, where it is stored with the previous K-1 instant flag values according to equations (3.21 ) and (3.22). iDFlag _buf(k) = iDFlag _buf(k - 1) k = K...2 (3.21) iDFlag _buf(ϊ) = iDFlag(n) (3.22)

3.9 Func5b -Tally Decision Flag (tPFIaq) and Instant Output Flag (iOFIaq)

The K most recent instant flag values stored in the "instant flag buffer" iDFIag_buf are processed in Func5b to produce a tally flag, tDFIag. Firstly, the flag buffer is analysed to find how many times each possible iDFlag value (0..6) occurs in the instant buffer, iDFIag_buf. This is performed according to equation (3.23).

tallyϋ) = ∑ fl(iDFlag _buf(k),i) i = 0..6 (3.23) k=\

The operator f 1 is defined in equation (3.24)

/KM) = I if(k = i)

(3.24) /1(M) = O else

A set of seven tally counts is produced, as shown in Table 3.8.

Table 3.8 Explanation of instant tally results. Then, the six instant tally counts corresponding to "known" states are analysed to find the flag value, tallyMax, whose tally value is both the maximum of the six and greater than a decision threshold, TyThreshi . This process is described by Table 3.9.

Table 3.9 TallyMax condition

The instant tally flag, tDFIag, is then set according to Table 3.10. This flag will only return an index to a "known" field ordering condition (O≤iDFlag <5) when the corresponding tally value is greater than the decision threshold, TyThreshi , and there are no other "known" or "error" conditions present in the buffer.

Table 3.10 Tally decision flag (tDFIag)

Funcδb also incorporates "instant intermittent" analysis, a procedure for detecting intermittent field misalignment in progressive content. Firstly, the progressive state that will potentially give visible errors, TestFlag, is determined according to Table 3.11 , where the IDFO (Interlaced Display Field Ordering) flag indicates whether the interlaced display is TFF or BFF. The IDFO flag indicates the supposed or alleged temporal relationship of the fields may be derived from the incoming signal, or may be input manually by the user.

Table 3.11 Progressive mis-alignment test flag (TestFlag) If IDFO signals TFF, then BFF misaligned progressive content might produce visible motion distortions and TestFlag is set to 0 (BFF). If IDFO signals BFF₁ then TestFlag is set to 2 (TFF). TestFlag is then used in a test of whether a significant number of potentially visible field misalignments have been detected in the instant flag buffer. The resulting instant output flag, iOFIag, is set according to Table 3.12. See Table 3.15 for threshold values.

Table 3.12 Instant output error flag (iOFIag)

3.10 Funcδc - Windowed Output Flags (wOFIaα. wEFIaq)

For a given buffer of frames, the "instant tally flag", tDFIag, reflects a condition where the majority of "instant" decision values have identical "known" conditions and all other decision values have the "unknown" condition. The flag, tDFIag, is combined with the windowed decision flag, wDFIag, according to Table 3.13 to produce the overall decision flag, wOFIag. This effectively copies the value of wDFIag to wOFIag if the instant tally flag is identical else an "unknown" condition is set.

Table 3.13 Combination of wDFIag and tDFIag

The output decision, wOFIag, indicates the frame properties (interlaced or progressive) and the field order properties for the buffered segment of content. Further processing is applied to indicate, through the error flag, wEFIag, whether the properties detected might cause visible errors for interlaced display. This function requires the additional input, IDFO (Interlaced Display Field Order) and is defined in Table 3.14. It should be noted that TFF misaligned progressive content output on a TFF interlaced display would not be flagged as an error and neither would BFF misaligned progressive on a BFF display.

Table 3.14 Error flag wEFIag

3.11 Output Flag Parameters (wOFIaq, wEFIaq. iOFIaq)

The output flags wEFIag and wOFIag reflect the results of the windowed analysis of field order properties. Parameters are averaged over a number of frames and refined by results from frame-by-frame "instant" results over the same period. As such, this analysis is sensitive to repetitive field-ordering conditions and may be considered to be a reliable detector of general content properties.

The output flag iOFIag is based on the analysis of frame-by-frame "instant" properties only and is sensitive to intermittent progressive field misalignments. The events handled by this flag are unlikely to be detected by the windowed analysis and the analysis therefore offers an additional warning for potentially visible errors. However, the nature of the "instant" test makes it more susceptible to misdetection and should be considered as a supplementary warning flag to the more reliable windowed results.

Further statistical analysis of these flags may be necessary to meet the requirements of differing applications.

3.12 Settings And Thresholds

The choice of decision thresholds will depend on the intended use of this field-order analysis algorithm. Table 3.15 shows a set of thresholds that suitable for the detection of potential field-ordering problems for standard resolution television systems.

Table 3.15 Setting and thresholds for Standard Resolution television

4 Algorithm Function and Performance There follows a consideration of function and performance of the key aspects of the field- detection algorithm described in section 3. Firstly, consideration is given in Sections 4.1 and 4.2 to the properties of "instant" and "windowed" decision parameters for interlaced and non-frame repeated progressive video. Then in Section 4.3 the benefits of combining "instant" and "windowed" parameters are presented.

4.1 Decision Parameter Properties For Interlaced Video

Considering Figure 12 for interlaced content, the expected time position of B(n) would be between T(n-1 ) and T(n) for BFF and between T(n) and T(n+1 ) f.or TFF. If there is significant motion between the fields, then field-ordering may be determined by a frame- by-frame comparison of the DIF values produced by Fund . Table 4.1 shows the expected DIF values for TFF and BFF interlaced video.

Table 4.1 Expected DIF conditions for interlaced content.

The DIF values for successive frames of a 7 second TFF interlaced sequence were calculated according to Fund and are shown in Figure 13. It can be seen that these "instant" frame values broadly satisfy the TFF test condition.

The noisy nature of the parameters may be handled by time-averaging and Figure 14 shows the content from Figure 13 with sliding 20-f rame window averaging applied according to Func2. Corresponding "instant" and "windowed" decision parameters calculated according to Func3 and Func3a are shown in Figures 15 and 16 respectively along with the corresponding interlaced decision thresholds, iThreshi and iThresh2.

Parameteri is intended to reflect the feature of two DIF values being nearly the same and significantly less than the remaining value and Parameter is intended to reflect the overall significance of the difference between the maximum and minimum DIF values. "Instant" and "windowed" decisions of interlaced type for the n'th frame are made by testing Parami and Param2 against thresholds, (func4) and (func4a). Successful detection of an interlaced type requires Parameteri to be below iThreshi and Parameter to be above iThresh2. The benefits of "windowing", particularly for parameter 1 , are clear from these graphs.

4. 2 Decision Parameter Properties For Progressive Content (non-frame repeated)

Considering Figure 12 for progressive content without frame repeats, field B(n) should be time-aligned with T(n). However, editing/processing might cause field B(n) to align with T(n-1 ) (BFF misaligned) or T(n+1 ) (TFF misaligned). If there is significant motion between successive frames, then field alignment may be determined by using the same DIF value comparisons as proposed for interlaced content. For such progressive content, Table 4.2 shows the expected DIF conditions for the progressive alignments.

Table 4.2 Expected conditions for progressive content.

Figure 17 shows "windowed" DIF values for 8 seconds of correctly aligned progressive content exhibiting the expected condition of (DIF2=0)«(DIF1=DIF3). Figure 18 shows "windowed" DIF values for progressive content with TFF misalignment, where (DIF3=0)«DIF2<DIF1.

The corresponding windowed parameter values, wParami and wParam2, for Figures 17 and 18 are shown in Figures 19 and 20 respectively.

For progressive content with significant motion between frames, a value of wParam2=0 clearly distinguishes both correctly and incorrectly aligned content from interlaced content. For correctly aligned progressive content, the value of WParami would be expected to be near 100% due to the small separation of the first and second ranked difference values. However, for misaligned progressive content, this feature is not present and an expected range of WParami >25% might be used to provide additional confidence to the discrimination. 4.3 Combining "Instant" and "Windowed" Parameters

The "windowed" decision parameters for both interlaced and non-frame-repeated progressive video are generally stable, offering reliable identification of field-ordering properties. However, events including scene-cuts, fast fades and fast pans may produce instant DIF values that make the "windowed" parameters temporarily unreliable.

Figure 21 shows the DIF parameters for a TFF interlaced sequence with a scene cut around frame 35. Figure 22 shows the equivalent "instant" parameter values and Figure 23 the "windowed" parameters. The key feature of this content is the scene cut at frame 35, which causes the distinct peaks in the Dl F1 and DIF3 measures. The scene cut is preceded by a period of low and reducing motion from frame 21 , which is reflected by all three DIF values in Figure 21 approaching zero.

The low motion of frames 27 to 31 fails to set the "instant" motion flag in Equation 3.14 and is indicated by the zeroing of both "instant" decision parameters in Figure 22. For interest this also occurs at frames 34, 67, 88, 89 and onwards from 97. Scene cuts have the property for "instant" analysis of returning an "unknown" classification rather than a "known" misclassification for the frames that are part of the cut. This short-lived effect can be observed in Figure 22. However, other than for the scene cut and the low motion frames, Figure 22 shows a fairly good "instant" classification of TFF interlaced content.

The advantage of "windowing" is clearly shown in Figure 23, with reliable and stable TFF classification up to frame 32 and beyond frame 53. However, a disadvantage is clear from frame 33 to 52, where the relatively large DIF values at the scene cut have distorted the "windowed" parameters for the 20 frame duration of the window. Table 4.3 shows this in detail, with "instant" and "windowed" DIF values shown side-by-side.

Table 4.3 Frame values for TFF interlaced scene cut

Table 4.3 also shows that, not only does the scene-cut distort the following "windowed" DIF values, but it also causes the "windowed" decision, wDFIag, to return a misclassification of "1=Correctly aligned progressive" for frames 35 onwards.

The field-order classification method described in Section 3 overcomes the problems of scene cuts, and more generally high motion fades and pans, by combining "instant" and "windowed" techniques. The "instant" parameters are more variable in nature, but far more robust to such conditions than the "windowed" parameters. Table 4.3 shows that the combined decision, wOFIag, rejects the misclassification from the "windowed" analysis, wDFIag.

Twenty minutes of TFF interlaced standard resolution television was analysed and the "windowed" and "combined" classification results are presented in Figure 24. It can be seen that the small proportion of misclassifications to progressive by the "windowed" analysis is removed by the "combined" analysis.

4. 4 Frame-repeated Progressive Content

Progressive content with repeated frames (Figure 4) presents the special problem that each bottom field tested may have identical time alignment with at least two top fields. Figure 25 shows "instant" DIF values for correctly aligned 2x frame-repeated progressive content. As each bottom field being tested is time-aligned to the corresponding top field and, alternately, the preceding and next top fields, then DIF2 is constantly near zero and DIF1 and DIF3 are saw-tooth in nature. Figure 26 shows instant DIF values for BFF misaligned 2x frame-repeated content (Figure 8). Here, for every frame DIFI=O and DIF1«DIF3 and for alternate frames DIF2=DIF1 and DIF2^«DIF3.

Such regular-frame repeating would require the "windowed" analysis described in Section 3 to be modified, as time-averaging the DIF values shown in Figures 25 and 26 according to equations 3.4 to 3.6 would not be sensible. If the nature of the frame-repeating was guaranteed to be regular and consistent, then techniques such as median-smoothing might be incorporated prior to "windowed" analysis. However, cartoon content has been found to often have time-varying frame-repeating properties with random field misalignments.

Rather than looking to classify all regions of such content, the technique described in Section 3 aims to identify only potentially visible misalignments. This is achieved within Funcδb by searching the "instant" flag buffer for potentially problematic patterns. This technique benefits from the high reliability of the "instant" decisions for progressive content.

Ten minutes of standard resolution progressive cartoon content with variable frame- repeating and BFF misalignment was analysed for a target TFF display. Figure 27 shows the classification results for combined analysis, wOFIag, and "instant intermittent" analysis, iOFIag. The "instant intermittent" test can be seen to add significantly to the sensitivity of the algorithm to irregular problem field misalignments. It is particularly useful for its ability to cope with varying degrees of frame-repeating, which is a condition that is not handled well by the "windowed" analysis.

5. Conclusion The technique described here has been designed to identify potential field-ordering problems for the interlaced display of progressive and interlaced content. A sliding 3- frame analysis window is used to generate "instant" and "windowed" sets of field difference measures, from which corresponding "instant" and "windowed" decision parameters are derived. These parameters are designed to reflect key identifying features of the field-ordering of progressive and interlaced video sequences. A combination of "windowed" and "instant" analysis has been shown to offer a reliable classification technique able to handle both interlaced and progressive sequences. The "combined" analysis makes the technique particularly robust to misclassification at scene- cuts, high motion pans and fades or with frame-repeated progressive content. Furthermore, the identification of potential problems in frame-repeated progressive content is enhanced by the inclusion of an "instant"-only test for irregular or intermittent problems.

6. References

[1] ITU-R Rec. BT.601 , "Studio encoding parameters of digital television for standard 4:3 and wide screen 16:9 aspect ratios," http://www.itu.int/rec/R-REC-BT.601/en.

[2] Kahn and Sanders, "Detection of progressive frames in a video field sequence," US Patent 6563550, May 13^th 2003.

[3] Casavant, Hurst, Perlman, lsnardi and Aschwanden, "Video/film-mode detector using patterns of 2-field differences", US Patent 5317398, May 1994.

[4] Westerman and Modem, "System for identifying video fields generated from film sources," US Patent 6157412, Dec 2000. [5] Honda and Nagakubo, "Method for identifying input video signal and processing the video signal," US Patent 6154257, Nov 2000.

[6] Swartz, "Film source video detection," US Patent 6014182, Jan 2000.

[7] Selby, "Method and apparatus for detecting motion between odd and even fields," US Patent 6633612, Oct 2003. [8] Faroudja, Xu and Swartz, "Motion detection between even and odd fields within 2:1 interlaced television standard," US Patent 5291280, March 1994.

[9] Christopher and Correa, "Method and Aparatus for Identifying ..," US Patent 5689301 , Nov 1997.

Claims

1. A method of detecting field order of a video signal, comprising: receiving successive digitally coded frames, each frame comprising data for a field of a first type and data for a field of a second type; generating for each field of the first type:

- a first difference signal (Dl F1 ) representative of the difference between the field and the second-type field of the previous frame;

- a second difference signal (Dl F2) representative of the difference between the field and the second-type field of the same frame;

2. A method according to claim 1 including receiving a signal (IDFO) indicating a purported temporal relationship, comparing it with the decision signal, and generating an alarm signal (wEFIag) in the event of a mismatch.

3. A method according to claim 1 or 2 in which the decision signal depends also on difference signals obtained in respect of other fields of the first type.

4. A method according to claim 3 in which each difference signal is averaged over a plurality of frames to produce a respective averaged difference signal (wDIF1 , wDIF2, wDIF3) and the decision signal (wOFIag) is obtained by analysing the averaged difference signals.

5. A method according to claim 3 comprising generating for each field of the first type an instantaneous decision signal dependent on said difference signals and then generating a combined decision signal (tDFIag) dependent on the instantaneous decision signals over a plurality of frames.

6. A method according to claim 3 in which each difference signal is averaged over a plurality of frames to produce a respective averaged difference signal (wDIF1 , wDIF2, wDIF3) and a first decision signal (wDFIag) is obtained by analysing the averaged difference signals; and comprising generating for each field of the first type an instantaneous decision signal dependent on said difference signals and then generating a second decision signal (tDFIag) dependent on the instantaneous decision signals over a plurality of frames; and combining the first and second decision signals.