OpenDML AVI File Format Extensions

Version 1.02

OpenDML AVI M-JPEG File Format Subcommittee

Last revision: February 28, 1996

Reformatting: September 1997
Please note
The information provided by this document is provided on an "as is" basis. No responsibility is assumed
by Matrox Electronic Systems Ltd. for its use; nor for any infringements of patents or other rights of third
parties resulting from its use. No license is granted under any patents or patent rights of Matrox
Electronic Systems Ltd.
 i

Table of Contents

1.0 Introduction ............................................................................................................................... 1

Scope ...................................................................................................................................... 1
OpenDML Sub-Committee on File Format: High Level Goals .................................................. 1
Version History ....................................................................................................................... 1
1.0 Overview of Profession Video Requirements ............................................................................ 2
Goal ........................................................................................................................................ 2
Issues ...................................................................................................................................... 2
AVI: Support Current AVI Format ...................................................................................... 2
60 Fields Per Second vs. 30 Frames Per Second; or 24 Frames Per Second For Film............. 2
Improved Tombstone and Header Information ...................................................................... 2
2.0 Increased AVI File Size ............................................................................................................. 3
Introduction: RIFF Chunk Format............................................................................................ 3
Increased File Size Limits (>> 1 GB) ....................................................................................... 3
Extension for File Size > 1 GB ................................................................................................ 3
3.0 Frame and Field Indexing .......................................................................................................... 5
Introduction: ‘idx1’ Index Chunk ............................................................................................. 5
Issues ...................................................................................................................................... 5
Allow Field Indexing............................................................................................................ 5
Allow Incremental Growth of Files....................................................................................... 5
Minimize Disk Seeks ........................................................................................................... 5
Proposed Index List ................................................................................................................. 6
Overview ............................................................................................................................. 6
Base Index Form ‘indx’ ....................................................................................................... 6
AVI Standard Index Chunk .................................................................................................. 7
AVI Field Index Chunk ........................................................................................................ 8
AVI Super Index Chunk ...................................................................................................... 9
Index Locations in RIFF File ............................................................................................. 10
4.0 Interleaving Audio with Video ................................................................................................ 13
Issues .................................................................................................................................... 13
Interleaving Rate ............................................................................................................... 13

Table of Contents
ii

Single or Multiple Files for Video and Audio...................................................................... 13

Multiple Tracks of Audio Storage ...................................................................................... 13
Mono and Stereo Audio ..................................................................................................... 13
Audio / Video Time Stamping (Sync Point List) ................................................................. 13
Notes on Interleaving Audio and Video .................................................................................. 13
5.0 Source and Header Information Storage .................................................................................... 15
Goal ...................................................................................................................................... 15
Required Information............................................................................................................. 15
Main AVI Header (avih) .................................................................................................... 15
AVI Stream Header (strh) .................................................................................................. 15
AVI Stream Format (strf) .................................................................................................. 16
Extended AVI Header (dmlh) ............................................................................................. 16
Video Properties Header (vprp) .......................................................................................... 16
Source and Timecode Information ...................................................................................... 20
Film Transfer Log Information........................................................................................... 22
Microsoft-Defined Tombstone Data ....................................................................................... 23
6.0 Other File Issues ...................................................................................................................... 28
Issues .................................................................................................................................... 28
Continuation Over Multiple Files ....................................................................................... 28
Minimized Time to Open Files ........................................................................................... 28
Simultaneous Reading and Writing / Multiple Access ......................................................... 28
7.0 MJPG Motion JPEG DIB Extensions ..................................................................................... 30
Issues .................................................................................................................................... 30
ISO vs. Non ISO Formats .................................................................................................. 30
8-bit vs. 10-bit YUV .......................................................................................................... 30
Video vs. Bitmap ............................................................................................................... 30
Proposed Data Chunk Format ................................................................................................ 30
From the DIB spec:............................................................................................................ 30
And a new proposal: .......................................................................................................... 32
MJPEG Baseline DCT - YCbCr DIB Map ......................................................................... 33
Sector Alignment and Padding ........................................................................................... 34
8.0 OpenDML File Format Certification Procedures ................................................................... 36
Concurrently Develop a Test Suite to measure compatibility .................................................. 36

Table of Contents
 iii

Table of Contents
 1

1.0 Introduction
The OpenDML File Format Subcommittee is defining an AVI-compatible file format that
addresses the particular needs of professional video. As such, this document relates specific
proposals to these needs.
Further work by OpenDML will elaborate on extensions to the Codec and Vidcap
specifications of Video for Windows to meet the needs of professional video.

Scope
This document describes the proposed format of the OpenDML-compliant AVI extensions,
specifically for the Motion-JPEG DIB AVI file. This format is an extension to the AVI M-
JPEG DIB as defined in the JPEG DIB FORMAT technical note from Microsoft. The format is
based on the ISO 10918 defined JPEG format.
It is assumed that the reader is familiar with JPEG as defined in the ISO 10918 document. For
additional information on JPEG, see the ISO 10918 Information technology -- Digital
compression and coding of continuous-tone still images: Requirements and guidelines.
For additional information about the JPEG DIB format, see Microsoft Multimedia Technical
Note JPEG DIB FORMAT.
For additional information about RIFF files see the Microsoft Windows Software Development
Kit Multimedia Programmer’s Guide and Multimedia Programmer’s Reference.
For additional information on installable compressors and decompressors, see the Video
Compression/Decompression Drivers technical note from Microsoft.
Portions of the above Microsoft documents have been reprinted with permission of Microsoft
Corp.

OpenDML Sub-Committee on File Format: High Level Goals

• Interoperability and AVI Compatibility
• Large File Support
• Frame and Field Indexing
• Interleaving Audio with Video
• Source and Header Information Storage
• OpenDML File Format Certification Code
• Microsoft Certification and Communication

Version History
Version 1.0 November 13, 1995 First Release
Version 1.1 December 18, 1995 Includes small changes from Microsoft

Scope
2

1.0 Overview of Profession Video Requirements

Goal
Provide interoperability between different hardware and software vendors’ Motion JPEG
codecs. Support Video for Windows AVI file format for MJPG video, while improving the
performance of these systems.

Issues
AVI: Support Current AVI Format
The goal is to support interoperability by standardizing on the AVI file format for Motion-
JPEG video. This implies a minimum compatibility to support the standard AVI file access
interface, as well as the vidcap and codecs that are part of Video for Windows. This means
support of the MJPG DIB specification, supporting the RIFF file format as well as the
standard AVI stream headers, allowing for optional audio interleaving, and supporting one
storage of one frame of video per ##dc data chunk in the LIST ‘movi’ chunk of the AVI file.

60 Fields Per Second vs. 30 Frames Per Second; or 24 Frames Per

Second For Film
Professional Video Applications require the ability to sequence individual fields of the video
data for certain playback rates and video effects. For example, in order to do a slow motion
effect, it is not sufficient to just repeat the frames; the fields must be individually repeated. For
example, as sequence of fields in a file numbered 123456... should be played at half speed as
11223344... rather than 12123434.... As such, the individual fields should be accessible. AVI
requires that frames are stored per data chunk. Storing individual fields in each chunk would
break AVI applications.
The extended AVI format will allow for access to individual fields. In addition, the format will
allow for storage of 24 frame based files, such that file readers can properly convert between
24 frame-based data storage and 25 or 30 frame PAL and NTSC playback.

Improved Tombstone and Header Information

Professional video applications need extended information such as starting timecode and reel
ID to be contained inside the AVI file. More advanced information such as the location of
timecode discontinuities, etc. are also desirable.

OpenDML AVI File Format Extensions

 3

2.0 Increased AVI File Size

Introduction: RIFF Chunk Format
The AVI file begins as a standard RIFF chunk (note that this is a non-standard representation
that shows the 4-byte size):
RIFF (size) ‘AVI ’ ...

Standard RIFF AVI form

RIFF chunk data

RIFF
size
AVI

size

size
ID

ID
chunk size

There is only one RIFF chunk per file. RIFF sub-chunks may be either LIST chunks or regular
sub-chunks. The LIST chunk obeys the same structure and may have regular or LIST sub-
chunks; all other RIFF sub-chunks have just an ID identifier and a size. These regular sub-
chunks may not have sub-chunks.

Increased File Size Limits (>> 1 GB)

The current RIFF file format implies a maximum chunk size of 4 GB because the size is stored
as a 32-bit value. However, limitations to the RIFF parser code and MCIAVI limits the file to
only 1 GB. At data rates of 10 MB/s, a 1 GB file will last less than 2 minutes.

Extension for File Size > 1 GB

An AVI file can be extended beyond 1 GB by placing more than one RIFF chunk in the same
file (which is legal). Each RIFF chunk can have a maximum size of 1 GB. Standard AVI
applications will see the first RIFF chunk (RIFF ‘AVI”) as a standard AVI file, and this chunk
should be complete. Information that the file is extended over 1 GB should be placed in the
Extended AVI header (see below). Subsequent RIFF chunks will be identified by the AVIX
(for AVI extended) chunk id. These RIFF ‘AVIX’ chunks do not need to contain further data.
It is expected that the AVIX chunks only contain LIST ‘movi’ data. All other AVI information
should be stored in the first (RIFF ‘AVI’) chunk so that it is accessible by all applications.

Introduction: RIFF Chunk Format

4

RIFF AVI form for files > 1 GB

first ~1 GB of RIFF data extended chunk data

real file length

AVIX
RIFF

RIFF
in samples
size

size
AVI
hdrx
size

size

size
ID

ID
header indicating Extended chunk size
presence RIFF AVIX chunks
First chunk (< 1 GB)

OpenDML AVI File Format Extensions

 5

3.0 Frame and Field Indexing

Introduction: ‘idx1’ Index Chunk
The AVI file format specifies an optional ‘idx1’ chunk, which contains a list of the offsets and
chunk types of every chunk inside the LIST ‘movi’ chunk. This list is used to make AVI
playback and seeks more efficient because the location of each frame of video can be found
without scanning through each sub-chunk of the LIST ‘movi’ data.

Standard idx1 form

movi LIST movi chunk data idx1 chunk data
LIST

idx1
size

size
*
01dc

01dc
data

data
size

size
...

* data of idx1 chunk is an index to each movi

sub-chunk, includes offset and length of each chunk

Issues
Allow Field Indexing
In order to support field accurate indexing while still retaining AVI compatibility, the video
data must be stored one frame per ##dc ‘movi’ chunk, while an index must be present that
gives the locations of both fields (if they exist) within the frame. Applications could then use
field-based codecs to perform field-based effects (e.g., slow motion, etc.).

Allow Incremental Growth of Files

The ‘idx1’ chunk follows the LIST ‘movi’ chunk in the file. As such, it must be moved in order
to insert new data into the file. An interleaved frame index would be more efficient for the
purposes of file growth.

Minimize Disk Seeks

In playback, there are several causes of inefficiency. One is that the index is at the end of the
file, requiring a large seek when opening the file. The other is that the index itself contains all
the chunks inside the ‘movi’ data, not just those for one AVI stream. As such, the index must
be preprocessed so that the location of each frame in the index can be found.

Introduction: ‘idx1’ Index Chunk

6

Proposed Index List

Overview
Instead of using an index chunk like ‘idx1’, a new index structure is proposed. The index has
the following main characteristics:
• The can either be a single index chunk, or a two tiered index, with a super index pointing to
interleaved index segments in the ‘movi’ data.
• The index can include locations of the fields within a frame.
• The index has no restrictions to 4 GB file size.
• There is an index per data stream (if necessary).
The advantages of the index are as follows:
• The index can contain the locations of the individual fields in each chunk as well as the
chunk information as a whole. The ‘idx1’ chunk only contains indices of the frames.
• The index segments are interleaved within the ‘movi’ data, meaning that the file can be
grown. The ‘idx1’ chunk is present at the end of the file. It must be moved for the file to
grow.
• The index segment is smaller than an ‘idx1’ chunk, so the amount of data needed to be read
in order to access a particular frame is less. Also, if the index is interleaved before its data
(i.e., less efficient in write), then playback will be more efficient than the ‘idx1’ chunk.
• The index lists are present per stream number (##). As such, the list can be accesses like an
array, unlike the ‘idx1’ chunk, which contains entries for all ‘movi’ chunks from all
streams.
The index is implemented as an index of indexes (optional) and an index of chunks. Both are
based on a base index form.

Base Index Form ‘indx’

Thus the actual implementation is based on a base index form ‘indx’:
struct _aviindex_chunk {
FOURCC fcc;
DWORD cb;
WORD wLongsPerEntry; // size of each entry in aIndex array
BYTE bIndexSubType; // future use. must be 0
BYTE bIndexType; // one of AVI_INDEX_* codes
DWORD nEntriesInUse; // index of first unused member in aIndex array
DWORD dwChunkId; // fcc of what is indexed
DWORD dwReserved[3]; // meaning differs for each index
// type/subtype. 0 if unused
struct _aviindex_entry {
DWORD adw[wLongsPerEntry];
} aIndex[ ];
};

OpenDML AVI File Format Extensions

 7

The actual size of the entries of the aIndex array is entered into wLongsPerEntry.
Between the cb field and wLongsPerEntry the actual size of the array is known. The field
nEntriesInUse allows a chunk to be allocated longer than the actual number of used
elements in the array.
The types defined for the bIndexType and bIndexSubtype field are the following:
// bIndexType codes
//
#define AVI_INDEX_OF_INDEXES 0x00 // when each entry in aIndex
// array points to an index chunk
#define AVI_INDEX_OF_CHUNKS 0x01 // when each entry in aIndex
// array points to a chunk in the
file
#define AVI_INDEX_IS_DATA 0x80 // when each entry is aIndex is
// really the data
// bIndexSubtype codes for INDEX_OF_CHUNKS
//
#define AVI_INDEX_2FIELD 0x01 // when fields within frames
// are also indexed

AVI Standard Index Chunk

The AVI Standard Index chunk contains information that indexes AVI frames.
typedef struct _avistdindex_chunk {
FOURCC fcc; // ’ix##’
DWORD cb;
WORD wLongsPerEntry; // must be sizeof(aIndex[0])/sizeof(DWORD)
BYTE bIndexSubType; // must be 0
BYTE bIndexType; // must be AVI_INDEX_OF_CHUNKS
DWORD nEntriesInUse; //
DWORD dwChunkId; // ’##dc’ or ’##db’ or ’##wb’ etc..
QUADWORD qwBaseOffset; // all dwOffsets in aIndex array are
// relative to this
DWORD dwReserved3; // must be 0
struct _avistdindex_entry {
DWORD dwOffset; // qwBaseOffset + this is absolute file offset
DWORD dwSize; // bit 31 is set if this is NOT a keyframe
} aIndex[ ];
} AVISTDINDEX, * PAVISTDINDEX;

A single standard index chunk can only index data within a 4 GB region. The dwOffset
field points to the start of the data itself, and not to the start of the RIFF chunk for that field.

Proposed Index List

8

Standard Index Chunk

LIST movi chunk data

movi
LIST
size

01dc

01dc
data

data
ix01
size

size

size
* ...

* data of Standard AVI index chunk is an index to

each movi chunk data, includes offset and length of chunk.

The index chunk above is shown in the ‘movi’ data, but may be found in the stream header (see
below).

AVI Field Index Chunk

The AVI Field Index Chunk is the same as the Standard Index Chunk except that it contains
the locations of each field in the frame.
typedef struct _avifieldindex_chunk {
FOURCC fcc; // ’ix##’
DWORD cb;
WORD wLongsPerEntry; // must be 3 (size of each entry in
// aIndex array)
BYTE bIndexSubType; // AVI_INDEX_2FIELD
BYTE bIndexType; // AVI_INDEX_OF_CHUNKS
DWORD nEntriesInUse; //
DWORD dwChunkId; // ’##dc’ or ’##db’
QUADWORD qwBaseOffset; // offsets in aIndex array are relative to this
DWORD dwReserved3; // must be 0
struct _avifieldindex_entry {
DWORD dwOffset;
DWORD dwSize; // size of all fields
// (bit 31 set for NON-keyframes)
DWORD dwOffsetField2; // offset to second field
} aIndex[ ];
} AVIFIELDINDEX, * PAVIFIELDINDEX;

The field wLongsPerEntry is set to 3 because of the addition of the dwOffsetField2

field in the array. This is indicated by setting the sub type to AVI_INDEX_2FIELD.

OpenDML AVI File Format Extensions

 9

Field Index Chunk

LIST movi chunk data

movi
LIST
size

01dc

01dc
data

data
ix01
size

size

size
* ...

* data of Field Index Chunk is an index to each movi

sub-chunk data, includes offset and length of chunk,
as well as offset to the second field.

The index chunk above is shown in the movi data, but may be found in the stream header (see
below).

AVI Super Index Chunk

The Super Index Chunk is an index of indexes and is always found in the ‘indx’ chunk of an
AVI file. It is defined as follows:
typedef struct _avisuperindex_chunk {
FOURCC fcc; // ’ix##’
DWORD cb; // size of this structure
WORD wLongsPerEntry; // must be 4 (size of each entry in aIndex array)
BYTE bIndexSubType; // must be 0 or AVI_INDEX_2FIELD
BYTE bIndexType; // must be AVI_INDEX_OF_INDEXES
DWORD nEntriesInUse; // number of entries in aIndex array that
// are used
DWORD dwChunkId; // ’##dc’ or ’##db’ or ’##wb’, etc
DWORD dwReserved[3]; // must be 0
struct _avisuperindex_entry {
QUADWORD qwOffset; // absolute file offset, offset 0 is
// unused entry??
DWORD dwSize; // size of index chunk at this offset
DWORD dwDuration; // time span in stream ticks
} aIndex[ ];
} AVISUPERINDEX, * PAVISUPERINDEX;

The bIndexSubType is set to the type of index that the Super Index points to. If the index
chunks are Standard Index Chunks, then the value should be 0. If the index chunks are AVI
Field Index Chunks, then the value should be AVI_INDEX_2FIELD. This implies that a
stream cannot mix Field and Standard Index Chunks.

Proposed Index List

10

Super Index Chunk

movi
LIST

LIST
size

size
hdrl
dc dc

LIST

ix01

ix01
size

size

size
strl
* data * data

indx
size
** ... ...

indx
* data of ##ix chunk is an index to each movi
sub-chunk data, includes offset and length of chunk,
and possibly the offset to the second field.
** data of Super Index indx chunk is an index to each ##ix
movi sub-chunk.

If we use the ##pc chunks, then they must be indexed by their own Index List: Palette changes
should not appear in the same Index List as ##dc chunks, since they eliminate the ability to
access an element (frame) directly without having to scan the list.

Index Locations in RIFF File

Unlike the ‘idx1’ chunk, a single index is stored per stream in the AVI file. An ‘indx’ chunk
follows the ‘strf’ chunk in the LIST ‘strl’ chunk of an AVI header. This ‘indx’ chunk may
either be an index of indexes (super index), or may be an index to the chunks directly. In the
case of video, this means that the chunk is either a AVISUPERINDEX or an
AVIFIELDINDEX/AVISTDINDEX.
If the ‘indx’ chunk is a standard or field index chunk (i.e., not an index of indexes) then the
stream has only one index chunk and there is none in the ‘movi’ data.
If the ‘indx’ chunk is a Super Index, then the corresponding index chunks are marked with
‘ix##’ in the ‘movi’ data. The ## is the stream number, the same as for the ##dc or ##wb.
The index chunks can be either standard or field index chunks.
A file can be easily grown if it has a standard index in the ‘indx’ chunk position. The chunk
can be moved to a new ‘ix##’ chunk, and a new super index can be inserted into the stream
header (‘indx’ position). New ‘ix##’ chunks can be added to grow the file.
Note the reversal of the ‘##ix’ to ‘ix##’. This is for AVI backward compatibility. Note that
INDEX_IS_DATA streams remain as ‘##ix’.
Here is a sample RIFF file in the format shown by the RIFFWALK utility:
RIFF (7F038718) ’AVI ’
LIST (0001084C) ’hdrl’
avih (00000038)

OpenDML AVI File Format Extensions

 11

LIST (000044E0) ’strl’

strh (00000038)
strf (00000428)
indx (00003FF8) <- video superindex
vprp (00000064) <- video property header
LIST (0000405E) ’strl’
strh (00000038)
strf (00000012)
indx (00003FF8) <- audio superindex
LIST (0000404C) ’strl’
strh (00000038) <- timecode stream...
strf (00000000)
indx (00003FF8) <- timecode superindex
LIST (00000060) ’odml’
dmlh (00000054) <- size still temp.
LIST (7F000000) ’movi’
01wb (00002dF0)
00db (00030004)
01wb (00002dF0)
00db (00030004)
... time passes
02ix (00003FF8) <- data chunk for timecode stream
... time passes
ix00 (00003FF8) <- partial video index
ix01 (00003FF8) <- partial audio index
.. etc ..
idx1 (00030000) <- original index (optional)
RIFF (7F038718) ’AVIX’

Proposed Index List

12

OpenDML AVI File Format Extensions

 13

4.0 Interleaving Audio with Video

Issues
Interleaving Rate
Files for Editing
Files for Playback

Single or Multiple Files for Video and Audio

Multiple Tracks of Audio Storage
Mono and Stereo Audio
Support of multiple streams of data exists in AVI. A proposal to recommend that audio be
stored as two mono streams should be examined. Not all applications would be able to play
this back in stereo, however.

Audio / Video Time Stamping (Sync Point List)

It is necessary to add time stamping information to account (and correct) for drift between
audio and video when using hardware that is not able to maintain a perfect synchronization
between audio and video (via a hardware connection) and when there exists a drift in the
crystal frequency on audio hardware.
This can be accomplished by adding a sync point list into the file. A sync point list identifies
two samples in different streams that are to be played at the same time. Thus, drift can be
corrected by identifying sync points and if necessary, correcting for the drift during playback.

Notes on Interleaving Audio and Video

With the exception of AVI files playing off CD-ROM drives, it is not necessary to interleave
the audio and video into REC chunks. In fact, with higher video and audio data rates, it is
probably more efficient to store the audio in large consecutive blocks. As such, the audio can
be interleaved in large blocks (not necessarily done on a frame basis).
Audio data chunks may or may not be in integral frames, or multiple of frames. The other
option would be a constant number of bytes. In addition, audio may be interleaved before or
after the corresponding video. This is just a tradeoff between record and playback
performance. Typically, saving audio chunks after video in record will give better record
performance, which is important because hard drives tend to be less performant during write
operations. However, during playback, multiple streams and cuts may be played, reducing the
drive performance.

Issues
14

OpenDML AVI File Format Extensions

 15

5.0 Source and Header Information Storage

Goal
The goal is to implement a standard way to store certain header and tombstone information in
an AVI file so that other applications can interpret and use this data. In many cases, the header
data is not uniformly set between different AVI files.

Required Information
Main AVI Header (avih)
Total Frames
The dwTotalFrames field indicates the size (number of frames) within the first RIFF
‘AVI’ chunk.

AVI Stream Header (strh)

Quality Information
AVI defines the quality of a clip by a DWORD value from 0 to 10000. This value seems
arbitrary. For MJPG files, a standard quality should be defined. The quality can be defined by
either of these three modes:
• constant average data rate
• constant Q-table
• lossless
Lossless is not part of the Baseline process. As such it requires the definition of a new Process.
Under baseline, lossless can be achieved by setting the Q-tables to unity1.
Under constant data rate, the desired data rate for the file should be specified (in kb/s). This,
plus the number of fields per frame, should be able to determine the Q-tables required to
compress the image.

1 While a lossless (or pseudo lossless) process can be defined by setting the Q-tables to unity,
the ISO specification of lossless JPEG compression is a different process. Lossless encoding
as defined by ISO does not use DCT transformed data. For a sample to be encoded, a
predictor is formed from the reconstructed values of up to three neighborhood samples. This
gives a prediction of the of the sample to be encoded. The prediction is subtracted from the
actual value of the sample and the difference is losslessly entropy coded using either Huffman
or arithmetic coding. As far as forming a type of pseudo lossless compression from a baseline
JPEG process with unity Q tables, there is the potential of some slight loss of data during the
DCT process due to arithmetic precision. Practically speaking, there is no real image quality
loss, but the word lossless implies a guarantee of reconstructing the source image on a bit by
bit basis.

Goal
16

Under constant Q-table, there are two possibilities. Either the absolute Q-tables are given (8x8
matrices) per component (in our case, 2), or a factor which is used to generate any Q-table of a
standard one. The constant Q-table file would of course have the same tables repeated per
field, as per the MJPG DIB spec. The advantage of a Q-factor is the reduction of header
information. However, tweaked Q-tables would no longer be possible.
Scale / Rate
The scale and rate parameters will define the correct ratio for different video standards. Known
ratio values are (rate/scale):
NTSC 30000/1001
PAL 25/1
rcFrame
The on-disk RECT coordinates are 16-bit values. The RECT structure in NT is a 32-bit
(LONG) for each coordinate, but the on-disk ones remain 16-bit.

AVI Stream Format (strf)

Height
The MJPG DIB spec lists that heights less than 288 are for single frames (one chunk) and
greater than 288 are for interleaved fields. However, there is no real connection between frame
height and interleaving. The biHeight parameter refers to the raw height of the frame
(interleaved or not). OpenDML codecs that operate in field by field modes will be passed this
value divided by 2 to get the field height.

Extended AVI Header (dmlh)

typedef struct {
DWORD dwTotalFrames;
} ODMLExtendedAVIHeader;

Total Frames
The dwTotalFrames field indicates the real size of the AVI file. Since the same field in the
Main AVI Header ‘avih’ indicates the size within the first RIFF ‘AVI’ chunk.

Video Properties Header (vprp)

The video properties header identifies video signal properties associated with a digital video
stream in an AVI file. This header attempts to address two main video properties:
• The type of video signal (PAL, NTSC, etc., as well as the resolution of the video signal).
• The framing of the compression within a video signal.
The parameters can be used to uniquely describe a video signal.
typedef struct {
DWORD VideoFormatToken;
DWORD VideoStandard;
DWORD dwVerticalRefreshRate;
DWORD dwHTotalInT;
DWORD dwVTotalInLines;

OpenDML AVI File Format Extensions

 17

DWORD dwFrameAspectRatio;
DWORD dwFrameWidthInPixels;
DWORD dwFrameHeightInLines;
DWORD nbFieldPerFrame;
VIDEO_FIELD_DESC FieldInfo[nbFieldPerFrame];
} VideoPropHeader;

Video Format Token

enum {FORMAT_UNKNOWN, FORMAT_PAL_SQUARE, FORMAT_PAL_CCIR_601,
FORMAT_NTSC_SQUARE, FORMAT_NTSC_CCIR_601,...} VIDEO_FORMAT;

The format token indicates that a known standard is defined for the following data fields.
Those fields must be filled, but their value can be expected to be the defined standard. If the
format is defined as FORMAT_UNKNOWN then the fields may contain special values. Known
tokens are defined in the table below:
Tok Stan Ref H- V- Frame F F Pixel
en dard res Tota Tot Aspect ra ra Aspect
h l al Ratio m m Ratio
Rat e e (derived
e W H )
id ei
th g
ht
NTS NTS 60 858 525 0x0004 72 48 2160:194
C C 0003 0 5 0
CCI (4:3)
R
601
NTS NTS 60 780 525 0x0004 64 48 1:1
C C 0003 0 0
SQU (4:3)
ARE
PAL PAL 50 864 625 0x0004 72 57 2160:230
CCI 0003 0 6 4
R (4:3)
601
PAL PAL 50 944 625 0x0004 76 57 1:1
SQU 0003 8 6
ARE (4:3)
Video Standard
enum {STANDARD_UNKNOWN, STANDARD_PAL, STANDARD_NTSC, STANDARD_SECAM}
VIDEO_STANDARD;

Defines standards such as NTSC, PAL etc. Implicitly defines vertical refresh rate.
Vertical Refresh Rate
Used when an unknown standard is specified. Normally, 60 for NTSC, and 50 for PAL.

Required Information
18

H-Total in T
Defines the horizontal total, in T (one luminance sample: pixel)
V-Total in Lines
Defines the vertical total, in lines.
Active Frame Aspect Ratio
The aspect ratio is stored as a DWORD value with a word each storing the x:y ratio. For
example, 1 to 1 is 0x00010001. Standard values for television is 4:3 or 16:9. This value can be
used with the frame width and height to calculate the pixel aspect ratio.
Active Frame Width in Pixels
Defines the active frame width in pixels. The bitmap might digitize a region that is smaller or
bigger than the active video width.
Active Frame Height in Lines
Defines the frame height in lines. The bitmap might digitize a region that is smaller or bigger
than the active video height.
Number of Fields Per Frame
One or two, depending on whether the video is interlaced or progressive.
Field Framing Information
The field framing information defines where the compressed image is with respect to the video
signal. The data is present for each field (and may be different).
Typedef struct {
DWORD CompressedBMHeight;
DWORD CompressedBMWidth;
DWORD ValidBMHeight;
DWORD ValidBMWidth;
DWORD ValidBMXOffset;
DWORD ValidBMYOffset;
DWORD VideoXOffsetInT;
DWORD VideoYValidStartLine;
} VIDEO_FIELD_DESC;

OpenDML AVI File Format Extensions

 19

x
X Origin: Leading edge of HSync

CompressedBMWidth

ValidBM
XOffset

ValidBMYOffset ValidBMWidth

V-Total VideoYValidStartLine

CompressedBMHeight ValidBMHeight

VideoXOffsetInT

H-Total

Compressed Bitmap Height and Width

The compressed bitmap height and width represent the size of the compressed image. For
JPEG, these values are multiples of 8.
Valid Bitmap Height and Width, X and Y Offset
The valid bitmap height, width and x and y offsets represent the size of the valid data within
the compressed bitmap. Because padding may be required when compressing, it is not
guaranteed that all the data within the compressed image is valid. Note that compressing
blanking is still valid. In the case where all the compressed bitmap comes from the video
signal, then the valid height and width are equal to the compressed height and width, and the
offsets are 0.
Valid X-Offset In T
The VideoXOffsetInT is used to locate the x position of the start of the valid bitmap with
reference to the video signal. The value is a measurement in units of T, which is one
luminance-sampling clock, from the leading edge of the horizontal sync pulse (CCIR 624-3).
Valid Y Start Line
The VideoYValidStartLine field is used to locate the line that the valid bitmap starts
on. This value will be different for each field. (CCIR 624-3).
A typical value for the Framing Information for NTSC CCIR 601 would be:
Vide Co Co Vali Vali Vali Va Vide Video

Required Information
20

o mpr mpr d d d lid oX Y

Typ esse esse BM BM BM B Offs Valid
e d d Hei Wid X M et In Start
BM BM ght th Offs Y T Line
Hei Wid et Off
ght th set
Fiel 248 720 248 720 0 0 122 20
d1
Fiel 248 720 248 720 0 0 122 283
d1

Source and Timecode Information

The following information is used to describe timecode and source information inside an AVI
file.
Base Timecode Structure
All SMPTE timecode information is stored in the following format:
typedef union _timecode {
struct {
WORD wFrameRate; // 0 is 30 drop. do we need other drop
// frame rates?
WORD wFrameFract; // fractional frame. full scale is always 0x10000
LONG lFrame;
};
__int64 qwAll;
} TIMECODE, *PTIMECODE;

This format allows for a frame, and a frame fractional value to be specified, with the frame
rate stored in the low word of the value. As the values are stored, two timecodes can be
compared and subtracted. For drop frame code, no lFrame values are skipped, so a drop
frame timecode of [Link];00 would be a lFrame value of 107892.
Timecode Discontinuity Table (Stream) ‘tcdl’
The following structure defines a timecode discontinuity table. This would be stored as a
stream itself. The table would either be a table alone in the ‘indx’ chunk of that stream, or if
the table is large, an index to the table could be stored with the table itself stored as a ##ix in
the ‘movi’ data.
#define FILM_SEQUENCE_NONE 0
#define FILM_SEQUENCE_AABBBCCDDD 1
#define FILM_SEQUENCE_AAABBCCCDD 2
// structure of a timecode discontinuity table. note that this fits
// within the definition of the structure of an index chunk
//
typedef struct _timecode_dl_chunk {

OpenDML AVI File Format Extensions

 21

FOURCC fcc; // ’##ix’

DWORD cb; // sizeof this structure less 8
WORD wLongsPerEntry; // must be 4 or more
BYTE bIndexSubType; // must be 0
BYTE bIndexType; // AVI_INDEX_IS_DATA
DWORD nEntriesInUse; // index of first unused entry in aIndex array
DWORD dwChunkId; // ’tcdl’ (timecode discontinuity list)
DWORD dwReserved[3]; // future, must be 0
struct _timecode_dl_entry {
QUADWORD qwTick; // time in terms of this streams’s tick rate
TIMECODE timecode; // timecode
DWORD dwUser; // timecode user data
struct {
DWORD ColorFrame : 4; // Which frame in color sequence
DWORD ColorSequence : 4; // Duration in frames of complete sequence
DWORD FilmFrame : 5; // Offset into pull-down sequence
DWORD FilmSequenceType : 3; // One of FILM_SEQUENCE_XXX defines
DWORD Reserved : 16; // Future use - set to 0
} Flags;
TCHAR szReelId[32]; // source id
} aIndex[ ];
};

The Flags structure can optionally specify color framing, and film pull-down information. If all
fields are 0 then no such optional information is present. Notice that there is no explicit flag for
drop frame NTSC since that information in already stored in the TIMECODE structure.
If ColorSequence is non-zero value, then ColorFrame contains a frame number from 0
to ColorSequence-1, which specifies the color framing of the frame specified in the
timecode structure.
If FilmSequenceType is not FILM_SEQUENCE_NONE, then FilmFrame contains the
offset into the pull-down sequence corresponding to the frame specified in the timecode
structure. For example, if FilmFrame is 1 and FilmSequenceType is
FILM_SEQUENCE_AABBBCCDDD then the referenced frame is a BB frame. Again, if
FilmFrame is 1 and FilmSequenceType is FILM_SEQUENCE_AAABBCCCDD then
the referenced frame is an AB frame.
Timecode Stream ‘time’
This structure defines a timecode value for every tick of the timecode stream
// structure of a timecode stream data chunk. for storing every
// event in a timecode stream rather than just the discontinuities
//
typedef struct _timecode_stream_chunk {
FOURCC fcc; // ’##ix’
DWORD cb; // sizeof this structure less 8
WORD wLongsPerEntry; // must be 3

Required Information
22

BYTE bIndexSubType; // must be 0

BYTE bIndexType; // AVI_INDEX_IS_DATA
DWORD nEntriesInUse; // index of first unused entry in aIndex array
DWORD dwChunkId; // ’time’ (timecode stream)
DWORD dwReserved[3]; // future, must be 0
struct _timecode_stream_entry {
TIMECODE timecode; // timecode
DWORD dwUser; // timecode user data
struct {
DWORD ColorFrame : 4; // Which frame in color sequence
DWORD ColorSequence : 4; // Duration in frames of complete sequence
DWORD FilmFrame : 5; // Offset into sequence
DWORD FilmSequenceType : 3; // One of FILM_SEQUENCE_XXX defines
DWORD Reserved : 16; // Future use - set to 0
} Flags;
} aIndex[ ];
};

Film Transfer Log Information

#define FILM_LOG_EVERTZ 1
#define FILM_LOG_FLEX 2
// structure of a film transfer log. Note that this fits
// within the definition of the structure of an index chunk
//
typedef struct _film_transfer_chunk {
FOURCC fcc; // ’##ix’
DWORD cb; // sizeof this structure less 8
WORD wLongsPerEntry; // must be 4 or more
BYTE bIndexSubType; // must be 0
BYTE bIndexType; // AVI_INDEX_IS_DATA
DWORD nEntriesInUse; // index of first unused entry in aIndex array
DWORD dwChunkId; // ’film’ (film transfer log)
DWORD dwReserved[3]; // future, must be 0
struct _film_transfer_header {
DWORD dwLogType; // Which type of FILM_LOG_XXX file format
DWORD dwHeaderSize; // Size of this _film_transfer_header structure
DWORD dwReserved[4];// Future use - set to 0
// DWORD dwUserDefined[]; // Optional user data
} FilmHeader;
// BYTE bTransferLog[]; // The actual transfer log begins here
};

The size of the FilmHeader is given by the dwHeaderSize. This allows for user defined
additions to the header, which appear at the end of the FilmHeader. The ASCII film

OpenDML AVI File Format Extensions

 23

transfer log begins immediately after the FilmHeader structure. The size of the ASCII data
can be determined from the dwHeaderSize of the header and the cb size of the entire
structure. The ASCII data should be padded with NULLs to the nearest DWORD structure
size.
The contents of the bTransferLog array is byte for byte the contents of the ASCII transfer
log file of the type given in the dwLogType field. For now, two types are defined:
FILM_LOG_EVERTZ: the detailed .FTL file format can be obtained from
Evertz Microsystems, Ltd.
3465 Mainway
Burlington, Ontario L7M 1A9, Canada
905-335-3700
FILM_LOG_FLEX: the detailed FLEX file format can be obtained from
Time Logic, Inc.
11992 Challenger Ct.
Moorpark, CA 93021
805-529-1155

Microsoft-Defined Tombstone Data

The following tombstone data has already been defined by Microsoft:
Additional AVI information following the ‘strl’ chunk in the AVI header LIST ‘hdrl’ chunk.
(from AVI RIFF form, Video for Windows SDK):
Additional Header Data
After the ‘strl’ chunk, some AVI files might have additional header data. Additional
information can include timecode chunks that apply to the whole file. The following timecode
four-character codes are useful in AVI files:2
Four-character Description
code

ISMP Indicates the chunk contains the SMPTE timecode of the

digitization starting point. The time is expressed as a zero-
terminated text string of the form HH:MM:[Link]. If
performing MCI capture in AVICAP, this chunk is
automatically set from the MCI start time.

IDIT Indicates the chunk specifies the time and date digitizing
commenced. This time is contained in an ASCII string of

2Note that these chunks are Ixxx chunks, and as such, should be found in an INFO chunk (see
comment below)?

Microsoft-Defined Tombstone Data

24

exactly 26 characters and has the format “Wed Jan 02

[Link] 1990\n\0”. The ctime and asctime functions can be
used to create strings in this format.

The ‘ISMP’ timecode chunk usually corresponds to the starting timecode copied from
the first sample of a captured sequence. Use this chunk to index the captured sample to
the original sample. For example, if you need to recapture a video sequence, you can
use the timecode to reposition the original sequence and restart capture from that
point.
The ‘IDIT’ timecode chunk usually corresponds to the time the sample was captured.
This timecode provides a reference to the age of a sample and creates a history if a
series of samples is captured.
The weakness with the above section is that the ISMP information does not include any reel
ID, so the effective timecode is meaningless.
Additional information, global to RIFF files (from Multimedia File Formats, Windows 3.1
SDK):
The INFO List Chunk
The “INFO” list is a registered global form type that can store information that helps
identify the contents of the chunk. This information is useful but does not affect the
way a program interprets the file; examples are copyright information and comments.
An “INFO” list is a “LIST” chunk with list type “INFO.” The following example
shows a sample “INFO” list chunk:

LIST(’INFO’ INAM("Two Trees"Z)

ICMT("A picture for the opening screen"Z) )

An “INFO” list should contain only the following chunks. New chunks may be
defined, but an application should ignore any chunk it doesn't understand. The chunks
listed below may only appear in an “INFO” list. Each chunk contains a ZSTR, or null-
terminated text string.
Chun Description
k ID

IAR Archival Location. Indicates where the subject of the file is archived.
L
IAR Artist. Lists the artist of the original subject of the file; for example,
T “Michaelangelo.”

ICM Commissioned. Lists the name of the person or organization that

S commissioned the subject of the file; for example, “Pope Julian II.”

ICM Comments. Provides general comments about the file or the subject of
T the file. If the comment is several sentences long, end each sentence with

OpenDML AVI File Format Extensions

 25

a period. Do not include new-line characters.

ICO Copyright. Records the copyright information for the file; for example,
P “Copyright Encyclopedia International 1991.” If there are multiple
copyrights, separate them by a semicolon followed by a space.

ICR Creation date. Specifies the date the subject of the file was created. List
D dates in year-month-day format, padding one-digit months and days with
a zero on the left; for example, “1553-05-03” for May 3, 1553.

ICR Cropped. Describes whether an image has been cropped and, if so, how
P it was cropped; for example, “lower-right corner.”

IDI Dimensions. Specifies the size of the original subject of the file; for
M example, “8.5 in h, 11 in w.”

IDPI Dots Per Inch. Stores dots per inch setting of the digitizer used to
produce the file, such as “300.”

IEN Engineer. Stores the name of the engineer who worked on the file. If
G there are multiple engineers, separate the names by a semicolon and a
blank; for example, “Smith, John; Adams, Joe.”

IGN Genre. Describes the original work, such as “landscape,” “portrait,”

R “still life,” etc.

IKE Keywords. Provides a list of keywords that refer to the file or subject of
Y the file. Separate multiple keywords with a semicolon and a blank; for
example, “Seattle; aerial view; scenery.”

ILG Lightness. Describes the changes in lightness settings on the digitizer

T required to produce the file. Note that the format of this information
depends on hardware used.

IME Medium. Describes the original subject of the file, such as “computer
D image,” “drawing,” “lithograph,” and so on.

INA Name. Stores the title of the subject of the file, such as “Seattle From
M Above.”

IPLT Palette Setting. Specifies the number of colors requested when digitizing
an image, such as “256.”

IPR Product. Specifies the name of the title the file was originally intended
D for, such as “Encyclopedia of Pacific Northwest Geography.”

ISBJ Subject. Describes the contents of the file, such as “Aerial view of
Seattle.”

ISFT Software. Identifies the name of the software package used to create the
file, such as “Microsoft WaveEdit.”

Microsoft-Defined Tombstone Data

26

ISHP Sharpness. Identifies the changes in sharpness for the digitizer required
to produce the file (the format depends on the hardware used).

ISR Source. Identifies the name of the person or organization who supplied
C the original subject of the file; for example, “Trey Research.”

ISRF Source Form. Identifies the original form of the material that was
digitized, such as “slide,” “paper,” “map,” and so on. This is not
necessarily the same as IMED.

ITC Technician. Identifies the technician who digitized the subject file; for
H example, “Smith, John.”

OpenDML AVI File Format Extensions

 27

Microsoft-Defined Tombstone Data

28

6.0 Other File Issues

Issues
Continuation Over Multiple Files
With restrictions to the maximum file format and limitations on the maximum size of disk
partitions, it may be desirable to implement a mechanism that specifies a linkage across
multiple files. It will certainly be desired that the data of one stream of video can be captured
across multiple files. For the moment, this will be external to the AVI file.

Minimized Time to Open Files

tbd

Simultaneous Reading and Writing / Multiple Access

tbd

OpenDML AVI File Format Extensions

 29

Issues
30

7.0 MJPG Motion JPEG DIB Extensions

Issues
ISO vs. Non ISO Formats
For interoperability between various hardware and software codecs, it is not sufficient that the
file format be standardized; the contents of the video data chunks must be standardized. In
order to comply with industry standards, the AVI MJPG specification and OMF 2.0 file format
specifications, the standard ISO Motion JPEG specification is proposed, with certain additions
to support efficient identification of the interleaved fields in the frame.
Non-ISO formats could be used, provided that they provide the same outward appearance as
the ISO data chunk. This would require different codecs to decompress and compress the data.

8-bit vs. 10-bit YUV

This document currently addresses 8-bit [Link] YUV as per the Microsoft specification.

Video vs. Bitmap

Issues between the fact that a bitmap is a pure rectangle, while video starts with half lines;
Motion JPEG compresses on multiple of 8 lines, so what is the frame of video that is
compressed vs. the real frame.
See the section of the Video Properties Header in Section 5.0.

Proposed Data Chunk Format

The data chunk format is defined as per the Microsoft JPEG DIB FORMAT Technical Note,
Type 2: Motion JPEG DIB.

From the DIB spec:

Type 2: Motion JPEG
Motion JPEG DIBs shall accommodate interchange formats which satisfy the "General
sequential and progressive syntax" (ISO 10918 Part 1, Annex B, Para. B.2). A set of
images of this type with compatible parameters can be placed in an AVI file to describe a
motion sequence. Frame headers for these DIBs shall be limited to those specified in Para
B.2.2 of the cited Annex B. These types are SOF0, SOF1, SOF2, SOF3, SOF9, SOF10
and SOF11. Of the types accommodated, this specification provides implementation only
for the Baseline Sequential DCT (SOF0).
This DIB type contains incomplete JPEG data (Abbreviated Format per ISO 10918) and is
not intended for stand-alone single image frame disk files. It may be used within RIFF files
and other contexts where it is appropriate to:
a. Decode an image without supplying the associated JPEG Huffman tables.
This presumes the codec has been properly pre-initialized prior to image
decode.

OpenDML AVI File Format Extensions

 31

b. Request encoder output of compressed image data absent embedded Huffman

Tables.
All motion JPEG data will use YCbCr encoding.
In an AVI sequence all JPEG frames will be key frames as this ensures that within the AVI
and Video for Windows architecture all frames will be directly and independently
addressable.
For optimal size and speed during playback of an AVI file the Huffman data used by
motion JPEG will be fixed and defined by this document. This will make the individual
frames of every motion sequence smaller and more efficient to play back. Also as all
sequences of motion images use the same Huffman data and color space it is much more
likely that motion data can be directly exchanged without re-compression. A definition of
the Huffman data will be provided in MMREG.H (which is listed at the end of this
document) as a byte string which can be concatenated onto the start of a motion JPEG
image to form a valid still JPEG image-
MJPGDHTSeg = { X’FF’, DHT, length, JPEG Huffman table
parameters }
Q-table data is present and may vary in every frame of a motion sequence to permit control
over the bandwidth of sequences that contain bursts of frames of varying levels of
complexity. The restart interval used during the compression process may also vary for
every frame.
Only the interleaved form of YCrCb images is supported for motion JPEG data. This
implies that only one SOS segment will be present in any particular motion JPEG image.
The following applies (again from MS DIB spec):
As in the JPEG DIB format the JPEG stream syntax is used for the image data with the
following constraints. The following JPEG marker codes SOI, DRI, DQT, SOF0, SOS
and EOI are expected (mandatory) in the image data chunk, and the constrained values
shown in the example below are mandatory for the image data within the AVI stream.
Any parameters in the SOF0 (frame) and SOS (start of scan) headers that are duplicated in
the BITMAPINFOHEADER for JPEG must be the same. This would include Sample
Precision, subsampling, number of components (as implied by JPEGColorSpaceID),
etc. The number of lines and samples per lines in the SOF0 segment and the width and
height defined in the format chunk must match the main AVI header width and height
values. All of these values are expected to remain the same for every image data chunk in
the AVI sequence.
Within the image data chunk two JPEG segments beginning with the SOI marker and
ending with the EOI marker are allowed to accommodate field-interleaved streams. There
is an APP0 marker immediately following the SOI marker that contains information about
the video image. Specifically, this allows the identification of the two fields that comprise
an interleaved frame. This APP0 marker is expected to have the first 4 bytes following the
length bytes set to the characters ’A’, ’V’, ’I’, ’1’. The next byte indicates which field the
JPEG data was compressed from and has an expected value of 0x01 for the FIRST JPEG
data segment and 0x02 for the SECOND segment, indicating the FIRST and SECOND
fields respectively. If the stream is not field interleaved then this value will be 0x00 and

Proposed Data Chunk Format

32

there will only be one JPEG segment. The remaining seven bytes are expected to be set to
0 and will be ignored by the codec.
If a codec cannot handle the interleaved fields, the codec will use only the FIRST field and
will replicate the lines as necessary to provide an image that conforms to the image size
defined in the main AVI header. Conversely if a capture system only accesses a single field
of each source frame only a single field image may be present in a JPEG stream. This
implies that the single field data should be used as the source of both fields by a
decompressor that wishes to process full-interlaced data.

And a new proposal:

The APP0 marker will have an added use beside the field polarity and will also specify the
size of the current field (entirely, from SOI to EOI), as well as any padding required. This
is to allow applications and codecs to avoid being forced to scan for the second SOI-EOI
pair. The size information in the APP0 marker does not include any information about a
second field in the ##dc chunk. This way, fields can be copied as integral units without
having to have the contained data modified for the presence and size of neighbouring
fields.
The byte following the polarity will be reserved for future expansion. The remaining 8
bytes of this marker will be used to store two sizes values. The first value specifies the
length in bytes of the field (entirely, from SOI to EOI) including any padding after the EOI
(which should be small) and the second value specifies the size of the same field excluding
any padding after the EOI. Both size values will be stored in the usual ISO JPEG fashion
where bytes are stored in decreasing order with the most significant byte being first. The
length of this marker was previously 14 and is now set to 16.

MJPG DIB RIFF chunk (##dc)

SOI
##dc
SIZE

EOI
SOI
EOI

An example MJPG frame with two interleaved fields

First field (SOI EOI pair)

Second field (SOI EOI pair)
Padding

OpenDML AVI File Format Extensions

 33

MJPEG Baseline DCT - YCbCr DIB Map

A Baseline DCT - YCbCr will now therefore have the following look :
SOI (xFFD8)
APP0 (xFFE0),
Length (16),
"AVI1", Polarity, 0, FieldSize,
FieldSizeLessPadding.
DRI (xFFDD),
Length (4),
Restart interval.
DQT (xFFDB),
Length (132),
Precision (0), Table ID (0),
DQT data [64] for table 0,
Precision (0), Table ID (1),
DQT data [64] for table 1.
SOF0 (xFFC0),
Length (17),
Sample Precision (8),
Number of lines (biHeight : multiple of 8),
Sample per line (biWidth : multiple of 16),
Number of components (3)
ID of 1st component (1),
Sampling ratio of 1st component (H,V),
Q-table ID of 1st component (0),
ID of 2nd component (2),
Sampling ratio of 2nd component (H,V),
Q table ID of 2nd component (1),
ID of 3rd component (3),
Sampling ratio of 3rd component (H,V),
Q table ID of 3rd component (1).
SOS (xFFDA),
Length (12),
Number of components (3)
ID of 1st component (1),
DC and AC Huffman ID for 1st component (0,0),
ID of 2nd component (2),

Proposed Data Chunk Format

34

DC and AC Huffman ID for 2nd component (1,1),

ID of 3rd component (3),
DC and AC Huffman ID for 3rd component (1,1),
Start of spectral (0),
End of spectral (63),
Successive approximation bit position, high and
low (0,0).
*** IMAGE DATA ***
There could be some Restart markers.
Values are from xFFD0 to xFFD7.
EOI (xFFD9).
Be sure to have a single DQT as the previous ones will be overridden.
The order of the markers may affect some codecs.
IDs for the 3 component as recommended by the Microsoft DIB specification are 1,2 and
3.

Sector Alignment and Padding

The ##dc chunks may be sector aligned if necessary. As such, a padding chunk would be
inserted following the ##dc to preserve the RIFF format. Padding between fields should be
discouraged; however the ISO specification allows padding if the data contains 0xFF for
each padded byte.

OpenDML AVI File Format Extensions

 35

Proposed Data Chunk Format

36

8.0 OpenDML File Format Certification Procedures

Concurrently Develop a Test Suite to measure compatibility
Test older Codecs with new AVI files.
Test old AVI files with new Codecs.
Circulate all files through industry (i.e., everybody tests everybody else’s files).
File checker routine to verify AVI format. Program may grow to include test decompressions
using software Codec.
Test using MCIAVI with standard codecs.
Go to MS porting labs for specific video file format / Codec test week.

OpenDML AVI File Format Extensions