Network Working Group
Request for Comments: 1314
A. Katz
D. Cohen
April 1992

A File Format for the Exchange of Images in the Internet

Status of This Memo

This document specifies an IAB standards track protocol for the Internet community, and requests discussion and suggestions for improvements. Please refer to the current edition of the "IAB Official Protocol Standards" for the standardization state and status of this protocol. Distribution of this memo is unlimited.


This document defines a standard file format for the exchange of fax-like black and white images within the Internet. It is a product of the Network Fax Working Group of the Internet Engineering Task Force (IETF).

The standard is:

** The file format should be TIFF-B with multi-page files

supported. Images should be encoded as one TIFF strip

per page.

** Images should be compressed using MMR when possible. Images

may also be MH or MR compressed or uncompressed. If MH or MR compression is used, scan lines should be "byte-aligned".

        ** For maximum interoperability, image resolutions should
           either be 600, 400, or 300 dpi; or else be one of the
           standard Group 3 fax resolutions (98 or 196 dpi
           vertically and 204 dpi horizontally).

Note that this specification is self contained and an implementation should be possible without recourse to the TIFF references, and that only the specific TIFF documents cited are relevant to this specification. Updates to the TIFF documents do not change this specification.

Experimentation with this file format specified here is encouraged.

1. Introduction

The purpose of this document is to define a standard file format for exchange of black and white images using the Internet. Since many organizations have already started to accumulate and exchange scanned documents it is important to reach agreement about an interchange file format in order to promote and facilitate the exchange and distribution of such documents. These images may originate from scanners, software, or facsimile (fax) machines. They may be manipulated by software, communicated, shared, duplicated, displayed, printed by laser printers, or faxed.

This file format provides for the uniform transfer of high quality images at a reasonable cost and with reasonable speed whether these files are generated by scanners, totally by software (e.g., text-to- fax, bitmap-to-fax, OCR, etc), or by fax. Also the intent of this document is to remain compatible with future moves to multi-level (i.e., gray-scale), higher resolution, or color images. The format proposed here is supported by both commercially available hardware and commercial and public domain software for most popular platforms in current use.

The file format for images is a totally separate issue from how such files are to be communicated. For example, FTP or SMTP could be used to move an image file from one host to another, although there are complications in the use of SMTP as currently implemented due to file size and the need to move binary data. (There is currently a proposal for removing these limitations from SMTP and in particular extending it to allow binary data. See reference [1].)

One major potential application of the communications format defined here is to allow images to be sent to fax machines using the Internet. It is intended that one or more separate companion documents will be formulated to address the issues of standardization in the areas of protocols for transmitting images through the Internet and the issues of addressing fax machines and routing faxes. Just as the exchange format is separate from the transmission mechanism, it is also separate from how hosts store images.

This document specifies a common exchange format; it does not require a host to store images in the format specified here, only to convert between the host's local image storage formats and the exchange format defined here for the purpose of exchanging images with other hosts across the network.

This standard specifies the use of TIFF (Tagged Image File Format, see below) as a format for exchange of image files. This is not a specific image encoding, but a framework for many encoding techniques, that can be used within the TIFF framework. For example, within TIFF it is possible to use MMR (the data encoding of CCITT Group 4 fax, see below), MH or MR (the data encodings of CCITT Group 3 fax), or other encoding methods.

Which encoding technique to use is not specified here. Instead, with time the encoding schemes used by most document providers will emerge as the de-facto standard. Therefore, we do not declare any as "the standard data encoding scheme," just as we do not declare that English is the standard publication language. (However, we expect that most document providers will use MMR in the immediate future because it offers much better compression ratios than MH or MR.)

Similarly, TIFF does not require that an image be communicated at a specific resolution. Resolution is a parameter in the TIFF descriptive header. We do suggest that images now be sent using one of a set of common resolutions in the interests of interoperability, but the format accommodates other resolutions that may be required by specialized applications or changing technologies.

Occasionally, image files will have to be converted, such as in the case where a document that was scanned at 400 dpi is to be printed on a 300 dpi printer. This conversion could be performed by the document provider, by the consumer, or by a third party. This document specifies neither who performs the conversion, nor which algorithms should be used to accomplish it.

Note that this standard does not attempt to define an exchange format for all image types that may be transmitted in the Internet. Nothing in this standard precludes it from being used for other image type such as gray-scale (e.g., JPEG) or color images but, for the purposes of standardization, the scope of this document is restricted to monochromatic bitmapped images.

The developers of this standard recognize that it may have a limited lifespan as Office Document Architecture (ODA) matures and comes into use in the Internet; ultimately the class of images covered by this standard will likely be subsumed by the more general class of images supported by the ODA standards. However, at present, there does not appear to be a sufficient installed base of ODA compliant software and the ODA standards are not fully mature. This standard is intended to fill the need for a common image transfer format until ODA is ready. Finally, we believe that it should be possible to automatically map images encoded in the format specified here into a future ODA-based image interchange format, thus providing a reasonable transition path to these future standards.

2. Relationship to Fax

Transmission of facsimile (fax) images over phone lines is becoming increasingly widespread. The standard of most fax machines in the U.S. is CCITT Group 3 (G3), specified in Recommendations T.4 and T.30 [2] and in EIA Standards EIA-465 and EIA-466. G3 faxes are 204 dots per inch (dpi) horizontally and 98 dpi (196 dpi optionally, in fine-detail mode) vertically. Since G3 neither assumes error free transmission nor retransmits when errors occur, the encoding scheme used is differential only over small segments never exceeding 2 lines at standard resolution or 4 lines for fine-detail. (The incremental G3 encoding scheme is called two-dimensional and the number of lines so encoded is specified by a parameter called k.)

CCITT Group 4 fax (G4) is defined by the T.400 and T.500 series of Recommendations as well as Recommendation T.6 [2]. It provides for 400 dpi (both vertical and horizontal) and is a fully two-dimensional encoding scheme (k is infinite) called MMR (Modified Modified READ, where READ stands for: Relative Element Address Designate). G4 assumes an error free transmission medium (generally an X.25 Public Data Network, or PDN). Because of this, G4 is not in widespread use in the U.S. today.

The traditional fax bundles together four independent issues:

(1) Data presentation and compression;
(2) Data transmission;
(3) Image input from paper ("scanning"); and
(4) Image output to paper ("printing").

   This bundling supports, for example, the high quality CCITT Group 4
   (G4) images (400x400 dpi) but only over X.25 public data networks
   with error correction,  and similarly it supports the mid-quality
   CCITT Group 3 (204x98 and 204x196 dpi) but only over phone voice
   circuits (the Switched Telephone Network, or STN) without error
   correction.  This bundling does not support the use of any other data
   transmission capabilities (e.g., FTP over LANs and WANs), nor
   asynchrony between the scanning and the printing, nor image storage,
   nor the use of the popular laser printers for output (even though
   they are perfectly capable of doing so).

In conventional fax, images are never stored. In today's computer network environment, a better model is:

(1) Images are scanned into files or created by software; (2) These image files are stored, manipulated, or communicated; (3) Images in a file are printed or displayed.

The only feature of the CCITT fax that should be used is the encoding technique (preferably MMR, but with MR or MH allowed) which may be implemented with a variety of fax-oriented chips at low cost due to the popularity of fax.

"Sending a fax" means both encoding (and decoding) the fax images as well as transmitting the data. Since the Internet ALREADY provides several mechanisms for data transmission (in particular, FTP for general file transmission), it is unnecessary to use the data transmission methods specified in the CCITT standard. Within the Internet, each fax image should be stored in a file and these files could be transferred (e.g., using FTP, SMTP, RPC-based methods, etc.).

Fax machines should be considered just as scanners and printers are, as I/O devices between paper and files; but not as a transmission means. Higher quality Group 4 images are thus supported at low cost, while enjoying the freedom to use any computerized file transfer and duplication mechanism, standard laser printers, multiple printing (possibly at multiple remote sites) of the same image without having to rescan it physically, and a variety of software for various processing of these images, such as OCR and various drawing programs. We should be able to interoperate with files created by fax machines, scanners, or software and to be able to print all of them on fax machines or on laser printers.

The CCITT Recommendations assume realtime communications between fax machines and do not therefore specify any kind of fax file format. We propose using TIFF [3] which seems to be emerging as a standard, for encapsulation of encoded images. Because they assume realtime communications, the CCITT fax protocols require negotiations to take place between the sender and receiver. For example, they negotiate whether to use two-dimensional coding (and with what k parameter) and what (if any) padding there is between scan lines.

In our approach, the image in the file is already compressed in a particular manner. If it is to be sent to an ordinary fax machine using a fax board/modem, that board will perform the negotiations with the receiving fax machine. In the cases where the receiver cannot handle the type of compression used in the file, it will be necessary to convert the image to another compression scheme before transmission. (Most fax cards seem to either store images using the default values of the parameters which are negotiated or in a format which can quickly be converted to this. With currently available hardware and software, any necessary format conversion should be easy to accomplish.)

In conventional fax, if the compression used for a particular image

   is "negative" (i.e., the compressed form is larger than the
   uncompressed form, something that happens quite frequently with
   dithered photographic images), the larger compressed form of the
   image is still sent.  If the images are first scanned into files,
   this problem could be recognized and the smaller, uncompressed file
   sent instead.  (Also, Recommendations T.4 and T.6 [2] allow for an
   "uncompressed mode."  Thus, lines which have negative compression may
   each be sent uncompressed.  However, very few G3 fax machines support
   this mode.)

3. Image File Format

Image files should be in the TIFF-B format which is the bi-level subclass of TIFF. TIFF and TIFF-B are described in reference [3], cited at the end of this document. Images should be compressed using MMR (the G4 compression scheme) because it offers superior compression ratios. However, images may also be compressed using MH or MR (the G3 methods). MMR offers much better compression ratios than these (which are used in G3 fax because of the lack of an error-free communications path).

TIFF-F, described in [4], is the proposed subclass of TIFF-B for fax images. However, since TIFF-F was intended for use with G3, it recommends against certain features we recommend. Specifically, it suggests not using MMR or MR compression (we recommend MMR and allow MR) and prohibits uncompressed mode (which we allow and suggest for some photographic images). Apart from these, the TIFF-F restrictions should be followed. (Complete compatibility between the format specified here and TIFF-F can only be guaranteed for MH compressed images.)

[NOTE: Aldus Corp., the TIFF Developer, considers fax applications to be outside the scope of mainstream TIFF since it is not a part of general publishing which is what TIFF was originally designed for. They specify the LZW [5] compression scheme rather than MMR. We, however, are concerned with the transmission and storage of images rather than publishing. Therefore, we are more concerned with compression ratios and compatibility with CCITT fax than Aldus is.]

TIFF itself allows for gray-scale and color images. Image files should be restricted to TIFF-B for now because most of the currently available hardware is bi-level (1 bit per pixel). In the future, when gray-scale or color scanners, printers, and fax becomes available, the file format suggested here can already accommodate it. (For example, though JPEG is not currently a TIFF defined compression type, work is currently underway for including it as such.)

[NOTE: In this document, we will use the term "reader" or "TIFF reader" to refer to the process or device which reads and parses a TIFF file.]

3.A. TIFF File Format

Figure 1 below (reproduced here from Figure 1 of reference [3]) depicts the structure of a TIFF file.

TIFF files start with a file header which specifies the byte order used in the file (i.e., Big or Little Endian), the TIFF version number, and points to the first "Image File Directory" (IFD). If the first two bytes are hex 4D4D, the byte order is from most to least significant for both 16 and 32 bit integers (Big Endian). If the first two bytes are hex 4949, the byte order is from least to most significant (Little Endian). In both formats, character strings are stored into sequential bytes and are null terminated.

The next two bytes (called the TIFF Version) must be 42 (hex 002A). This does not refer to the current TIFF revision number. The following 4 bytes contain the offset (in bytes from the beginning of the file) to the first IFD.

An IFD contains a 2 byte count of the number of entries in the IFD, a sequence of 12 byte directory entries, and a 4 byte pointer to the next IFD. One of these fields (StripOffsets) points to (parts of) an image in the file. There may be more than one image in the file (e.g., a "multi-page" TIFF file) and therefore more then one IFD. IFD field entries may appear in any order.

Each directory entry is 12 bytes and consists of a tag, its type, a length, and an offset to its value. If the value can fit into 4 bytes (i.e., if the type is BYTE, SHORT, or LONG), the actual value rather than an offset is given. If the value is less than 4 bytes (i.e., if the type is BYTE or SHORT), it is left-justified within the 4 byte value offset. More details about directory entries and the possible tags will be given in Section 3.C.

All pointers (called offsets in the TIFF reference [3]) are the number of bytes from the beginning of the file and are 4 bytes long. The first byte of the file has an offset of 0. In the case of only one image per file, there should therefore be only one IFD. The last IFD's pointer to the next IFD is set to hex 00000000 (32 bits).

The entries in an IFD must be sorted in ascending order by Tag.

        +--------+--------+                     Directory Entry
      0 |        |        | Byte Order        +--------+--------+
        +--------+--------|               X   |        |        | Tag
      2 |        |        | Version(42)       +--------+--------|
        +--------+--------|               X+2 |        |        | Type
      4 |        |        | Offset of         +--------+--------|
        +-     - A -     -+  0th IFD      X+4 |        |        |
      6 |        |        |                   +-               -+ Length
        +--------+--------+                   |        |        |
                 |                            +--------+--------+
                 |                        X+8 |        |        | Value
                 |                            +-     - Y -     -+   or
                 V                            |        |        | Value
                                              +--------+--------+ Offset
        +--------+--------+                            |
  A     |      - B -      | Entry Count                |
        +--------+--------|                            |
        |        |        |                            V
  A+2                       Entry 0
        |        |        |                   +--------+--------+
        +--------+--------+                   |        |        |
        |        |        |                 Y                     Value
  A+14                      Entry 1           |        |        |
        |        |        |                   +--------+--------|
        |        |        |
  A+26                      Entry 2
        |        |        |
        |        |        |    .
        |        |        |    .
        |        |        |
                             Entry B-1
        |        |        |
        |        |        |  Offset of
A+2+B*12       - C -      +  Next IFD
        |        |        |
            (next IFD)

Figure 1: The Structure of a TIFF File

3.B. Image Format and Encoding Issues

Images in TIFF files are organized as horizontal strips for fast access to individual rows. One can specify how many rows there are in each strip and all of the strips are the same size (except possibly the last one). Each strip must begin on a byte boundary but successive rows are not so required. For two-dimensional G3 compression (MR), each strip must begin with an "absolute" one- dimensional line. For MMR (G4) compression, each strip must be encoded as if it were a separate image.

For a variety of reasons, each page must be a single strip (e.g., not broken up into multiple strips).

One problem with multiple strips per page is that images which come from G4 fax machines as well as most scanned images will be generated as a single strip per page. These would have to be decoded and re- encoded as multiple strips (remember that for MMR compression, each strip must be start with a one-dimensionally encoded line).

Another problem with multiple strips per page arises in MR compression. Here, there MAY be at most k-1 two-dimensionally encoded lines following a one-dimensionally encoded line, but this is not required. It is possible to have one-dimensional lines more frequently than every k lines. However, since each strip (except possibly the last one) is required to be the same size, it may be necessary to re-encode the image to insure that each strip starts with a one-dimensional line. This is not a problem if each page is a single strip.

[NOTE: The TIFF document [3] suggests using strips which are about 8K bytes long. However, TIFF-F [4] recommends that each page be a single strip regardless of its size. The format specified in this document follows the TIFF-F recommendation.]

   Also, as TIFF-F recommends, all G3 encoded images (MH and MR) should
   be "byte-aligned."  This means that extra zero bits (fill bits) are
   added before each EOL (end-of-line) so that every line starts on a
   byte boundary.

In addition, as in the TIFF-F specification, the RTC (Return to Control signal which consists of 6 continuous EOL's) of G3 shall not be included at the end of G3 encoded documents. RTC is to be considered part of the G3 transmission protocol and not part of the encoding. Most, if not all, G3 fax modems attach RTC to outgoing images and remove it from incoming ones.

For MMR (G4) encoded files, readers should be able to read images with only one EOFB (End Of Facsimile Block) at the end of the page and should not assume that Facsimile Blocks are of any particular size. (It has been reported that some MMR readers assume that all Facsimile Blocks are the maximum size.)

Systems may optionally choose to store the entire image uncompressed if the compression increases the size of the image file. Also, uncompressed mode (specified in Group3Options or Group4Options, see below) allows portions of the image to be uncompressed.

The multi-page capability of TIFF is supported and should be used for multi-page documents. TIFF files which have multiple pages have an IFD for each page of the document each of which describes and points to a single page image. (Note: though the current TIFF specification does not specifically prohibit having a single IFD point to an image which is actually multiple pages, with one strip for each page, most if not all TIFF readers would probably not be able to read such a file. Therefore, this should not be done.)


Since most publications (e.g., reports, books, and magazine articles) are composed of more than a single page, multi-page TIFF files should be used where appropriate. However, many current TIFF implementations now only handle single-page files.

It is hoped that in the future, more TIFF implementations will handle multi-page files correctly. In the meantime, it would be useful to develop a utility program which could join several single-page TIFF files into a single multi-page file and also separate a multi-page TIFF file into several single page files.

For example, the utility could take a single TIFF file
with N pages, called doc.tif, and create the files doc.000, doc.001, doc.002, ..., doc.N. doc.000 would be an ASCII listing of the files created. This naming
scheme is compatible with that used by the image systems we have seen which only handle single page files.

In going the other way, the N+1 single page files could be combined into a single multi-page TIFF file. In this case, if the file doc.000 exists but contains information contrary to what is found in looking for the files doc.001, doc.002, ..., the program would notify the user.]

3.C. TIFF Fields

TIFF is tag or field based. The various fields and their format are listed in [3]. There are Basic Fields (common to all TIFF files), Informational Fields (which provide useful information to a user), Facsimile Fields (used here), and Private Fields.

Each directory entry contains:

The Tag for the field (2 bytes)

The field Type (2 bytes)

The field Length (4 bytes)

(This is in terms of the data type, not in bytes. For

example, a single 16-bit word or SHORT has a Length

of 1, not 2)

The Value Offset (4 bytes)

(Pointer to the actual value, which must begin on a

word boundary. Therefore, this offset will always
be an even number. If the Value fits into 4 bytes, the Value Offset contains the Value instead. If the Value takes less than 4 bytes, it is left justified)

The allowed types and their codes are:

        1 = BYTE        8-bit unsigned integer (1 byte)
        2 = ASCII       8-bit ASCII terminated with a null (variable
        3 = SHORT       16-bit unsigned integer (2 bytes)
        4 = LONG        32-bit unsigned integer (4 bytes)
        5 = RATIONAL    Two LONGs (64 bits) representing the
                        numerator and denominator of a fraction.
                        In this document, RATIONAL's will be written
                        as numerator/denominator. (8 bytes)

For ASCII, the Length specifies the number of characters and includes the null. It does not, however, include padding if such is necessary.

(Note that ASCII strings of length 3 or less may be stored in the Value Offset field instead of being pointed to.)

The following fields should be used in a TIFF image file. Only the Basic Fields are mandatory; the others are optional (except that for MH and MR encoded files, the Group3Options Facsimile Field is mandatory). The optional fields have default values which are given in the TIFF specification. (Note that the TIFF reference [3] recommends not relying on the default values.)

Some fields contain one or more flag bits all stored as one value. In these cases, the bit labeled 0 is the least significant bit (i.e., Little Endian order). Where there is more than one suggested value for a tag, the possible values are separated by |.

Note that some fields (such as ImageLength or ImageWidth) can be of more than one type.

It would be useful to develop a TIFF viewer and editor which would allow one to read, add, and edit the fields in a TIFF file. Such an editor would display fields in sorted order and force the inclusion of all mandatory fields. Also, resolution and position should always be displayed or specified together with their units.

   3.C.1.  Basic Fields (Mandatory)

Basic Fields are those which are fundamental to the pixel architecture or visual characteristics of an image. The following Basic Fields should be included in a TIFF image file:

           FIELD NAME
       (TAG in hex, TYPE)       VALUE           DESCRIPTION
       ------------------       -----           -----------
         BitsPerSample            1             Number of bits
          (0102, SHORT)                         per pixel (bi-level for
                                                now, but may allow
                                                more later)
         Compression              4             Type of Compression
          (0103, SHORT)      (could also be       1 = Uncompressed
                                1 or 3)           3 = G3 (MH or MR)
                                                  4 = G4 (MMR)
                                                 Use 4 if possible

ImageLength <image's length> Length of the Image

(0101, SHORT in scan lines

or LONG)

ImageWidth <image's width> Width of the Image

(0100, SHORT in pixels

            or LONG)
         NewSubFileType     0 usually           Flag bits indicating
          (00FE, LONG)       bit 0: 1 if           the kind of image.
                              reduced           (see the TIFF
                              resolution of        reference [3])
                              another image
                             bit 1: 1 if
                              single page of a
                              multi-page image
                             bit 2: 1 if
                              image defines a
         Photometric-       0 for positive
           Interpretation    image (0 imaged
          (0106, SHORT)      as white, 1 as
                            1 means reverse
                             black and white
         RowsPerStrip    <Number of Rows>       Number of Rows in
          (0116, SHORT                          Each Strip.  Each
           or LONG)                             page should be a
                                                single strip.
         SamplesPerPixel          1             (since are Bi-level
          (0115, SHORT)                          images)
         StripByteCounts    count1, count2...   Number of Bytes in
          (0117, SHORTs                          each strip of the
            or LONGs)                            images.  (The Value
                                                 is an offset which
                                                 points to a series
                                                 of counts, each of
                                                 which is the same
                                                 Type, LONG or SHORT.
                                                 The Length is the
                                                 same as the number
                                                 of strips.)
         StripOffsets       off1, off2,...      Pointers to the strips
          (0111, SHORTs                          of the image (remember,
            or LONGs)                            one strip per page).
                                                 (The Value is an offset
                                                  which points to a
                                                  series of offsets,

each of which points to the actual image data for the strip.)

ResolutionUnit 2 | 3 Units of Resolution

(0128, SHORT) See Below, 3.C.6 2: Inches

3: Centimeters

         XResolution        See Below, 3.C.6    Resolution in the X
          (011A, RATIONAL)                       direction in pixels
                                                 per ResolutionUnit
                                                 (we suggest 400 dots
                                                 per inch when possible)
         YResolution        See Below, 3.C.6    Resolution in the Y
          (011B, RATIONAL)                        direction in pixels
                                                 per ResolutionUnit
                                                 (we suggest 400 dots
                                                 per inch when possible)
   3.C.2.  Informational Fields (Optional)

The following Informational Fields are optional. They provide useful information to a user. All Field values are ASCII strings.

       NAME (TAG in hex)                DESCRIPTION
       ----------------                 -----------

Artist (013B) Person Who Created the Image

DateTime (0132) Date and Time of Image Creation

         HostComputer (013C)     Name of Computer Image was Created On
         ImageDescription        A Short Text Description
         Make (010F)             Manufacturer of Hardware (Scanner) Used
         Model (0110)            Model Number of Hardware (Scanner) Used
         Software (0131)         Software Package that Created the Image
   3.C.3.  Facsimile Fields (Optional, Mandatory for G3 Compression)

In addition to the above, the Facsimile Fields below should be used. The TIFF document recommends that they not be used for interchange between applications, but they are now in wide enough use for just that. These fields are optional and default to 0 (all bits off).

           FIELD NAME
       (TAG in hex, TYPE)       VALUE               DESCRIPTION
       ------------------       -----               -----------
         Group3Options      bit 0: 1 for         Flag bits indicating
          (0124, LONG)       2-dimensional       Options for G3
                              (i.e., MR with
                               k > 1)
                            bit 1: 1 if
                             mode MAY be used,
                             0 if uncompressed
                             mode IS NOT used.
                            bit 2: 1 if fill     (As allowed by the G3
                             bits have been       protocol, fill bits
                             added                may be added between
                                                  each line of data
                                                  and the EOL.  Since
                                                  fill bits are used to
                                                  "byte-align" G3 image
                                                  files, bit 2 should be
                                                  set to 1 for these
         Group4Options      bit 0: unused        Flag bits indicating
          (0125, LONG)      bit 1: 1 if          Options for G4
                             mode MAY be used,
                             if this bit is 0
                             it means that
                             uncompressed mode
                             IS NOT used.
   3.C.4.  Storage and Retrieval Fields (Optional)

The following fields are optional and may be useful for document storage and retrieval.

           FIELD NAME
       (TAG in hex, TYPE)                DESCRIPTION
       ------------------                -----------
         DocumentName               Name of the Document
          (010D, ASCII)
         PageName                   Name of the Page
          (011D, ASCII)
         PageNumber                 Page Number in a Multi-Page Document
          (0129, SHORTs)             Two SHORT Values are specified, the
                                     first is the page number and the
                                     second is the total number of pages
                                     in the document.  The first page
                                     is page 0.  (NOTE:  This does not
                                     necessarily correspond to page
                                     numbers which may be printed
                                     in the image.)

XPosition X Offset of the Left Side of

(011E, RATIONAL) the Image, in ResolutionUnits

         YPosition                  Y Offset of the Top of
          (011F, RATIONAL)          the Image, in ResolutionUnits
   3.C.5.  TIFF-F Fields (NOT Recommended)

TIFF-F defines the following new fields for G3 (MH) encoded images. Since these fields are not defined in TIFF-B itself, their use is not recommended. However, since TIFF-F files may include these tags for image data which came from a G3 fax machine, readers should be prepared for them.

These three fields deal with corrupted image data which is due to the fact that G3 devices may not perform error correction on bad data.

           FIELD NAME
       (TAG in hex, TYPE)                DESCRIPTION
       ------------------                -----------

BadFaxLines Number of Bad fax scan lines

(0146, SHORT or LONG) encountered during fax reception

(but not necessarily in the file)

         CleanFaxData               0 means no bad lines received
          (0147, SHORT)             1 means bad lines were regenerated
                                        by the receiving device
                                    2 means bad lines were detected
                                        but not regenerated
        ConsecutiveBadFaxLines      The maximum number of consecutive
          (0148, SHORT or LONG)     bad fax lines (but not necessarily
                                    in the file)
   3.C.6.  More on Representing Resolutions

The tags XResolution and YResolution are both RATIONALs, i.e., the ratio of two LONGS. G3 fax resolutions are actually specified in dots (or lines) per mm while G4 is in dots per inch (actually, dots per 25.4 mm).

For example, G3 horizontal resolution is defined to be 1728 dots per 215 mm which comes out to 80.4 dots per cm or about 203 dots per inch. It is frequently referred to as just 200 dpi. To avoid any possibility of problems due to round off error, this should be represented by having XResolution = 17280/215 and ResolutionUnit = 3 (cm). However when reading, 204/1 or even 200/1 with ResolutionUnit = 2 (inches) should be recognized as representing the same resolution.

For G4, on the other hand, the resolution 400 dots/inch should be represented by an XResolution of 400/1 and ResolutionUnit = 2.

The following table shows various ways of representing the standard resolutions in order of preference:

                   ResolutionUnit    XResolution       YResolution
                   --------------    -----------       -----------
        G3 normal       3             17280/215         3850/100
                        3                80/1           3850/100
                        3             17280/215          385/10
                        3                80/1            385/10
                        2              2042/10          9779/100
                        2               204/1             98/1
                        2               200/1            100/1
        G3 fine         3             17280/215           77/1
                        3                80/1             77/1
                        2              2042/10         19558/100
                        2               204/1            196/1
                        2               200/1            200/1

G4 200 dpi 2 200/1 200/1

G4 300 dpi 2 300/1 300/1

Other 300 dpi 2 300/1 300/1

G4 400 dpi 2 400/1 400/1

600 dpi 2 600/1 600/1

It is suggested that Image readers be able to handle all of the above representations.

4. A Sample TIFF Image File

Below is a sample of what might be in a TIFF file for an MMR (G4) encoded single image which is about 100K bytes compressed at 400 dpi. A generic outline is given first, followed by a more detailed hex listing.

4.A. Sample File

Comments are to the right and are preceded by a semicolon. Note that tags must be sorted in order of the tag codes.

0:, IFDADDR:, and STRIP0: are addresses within the file and denote the number of bytes from the beginning of the file.


    0:  Byte Order=     hex 4D4D        ;first bytes of the file, from
                                        ;most significant bit to least
                                        ;significant (big endian)
        Version=        42 (hex 002A)   ;Must be 42
        First IFD=      IFDADDR         ;Address of first (and only) IFD

Image File Directory (the only one in this example):


        IFD Entry Count=      24        ;(NOT A TAG) Count of
                                        ; Number of IFD Entries
        NewSubFileType=        0
        ImageWidth=         3400        ;8.5 inches at 400 dpi
        ImageLength=        4400        ;11 inches at 400 dpi
        BitsPerSample=         1        ;Bi-Level
        Compression=           4        ;MMR
           Interpretation=     0
        DocumentName=       "LAMap1"
        ImageDescription=   "A map of Los Angeles"
        Make=               "Fujitsu"
        Model=              "M3093E"
        StripOffsets=       <STRIP0>    ;There is only one strip in
                                        ;this example.  However, note
                                        ;that strips can be in any
                                        ;order.  (Offsets are from the
                                        ;beginning of the TIFF file.)
        SamplesPerPixel=       1        ;Bi-Level
        RowsPerStrip=       4400        ;Entire image in 1 strip
        StripByteCounts=    <COUNT0>    ;Byte count of entire
                                        ;compressed image
        XResolution=        400/1
        YResolution=        400/1
        XPosition=            0/1       ;position of left side of image
        YPosition=            0/1       ;position of top of image
        Group4Options=    hex 00000002  ;bit 1 on means uncompressed
                                        ;mode MAY be used
        ResolutionUnit=        2        ;Inches
        Software=           "Xionics"
        DateTime=           "1990:10:05 15:00:00"
        Artist=             "Joe Pro"
        HostComputer=       "Tardis.Isi.Edu"
        Next IFD Pointer=  hex 00000000 ;(NOT A TAG) Indicates no
                                        ; more IFDs in this file

Image Data:

    <STRIP0>:       <actual compressed image data>

[end of TIFF file]

In this example there is only one strip. Note that if there were more than one, the TIFF specification does not require them to be in any particular order. Strips may be given in any order and TIFF readers must use the StripOffsets to locate them.

Also, the TIFF document recommends not relying on the default values of the tags.

4.B. Detailed Hex Listing

All offsets and values are represented by hex except for ASCII strings which are double quoted. Remember that Value Offsets must always be an even number since the value it points to must always be on a 16-bit word boundary.

Entries in the Name column are for reference and are not actually a part of the TIFF file.

   Offset      Name                  Value
   ----        -------------------   -------------------------------------
  Header (first byte is Offset 0):
   0000        Byte Order             4D4D
   0002        Version                002A
   0004        1st. IFD pointer       00000010

IFD (IFDADDR from above is 0010 here):

   0010        Entry Count            0018
   0012        NewSubFileType         00FE   0004   00000001  00000000
   001E        ImageWidth             0100   0004   00000001  00000D48
   002A        ImageLength            0101   0004   00000001  00001130
   0036        BitsPerSample          0102   0003   00000001  00010000
   0042        Compression            0103   0003   00000001  00040000
   004E        Photometric Interp.    0106   0003   00000001  00000000
   005A        DocumentName           010D   0002   00000007  00000136
   0066        ImageDescription       010E   0002   00000015  0000013E
   0072        Make                   010F   0002   00000008  00000154
   007E        Model                  0110   0002   00000007  0000015C
   008A        StripOffsets           0111   0004   00000001  000001A8
   0096        SamplesPerPixel        0115   0003   00000001  00010000
   00A2        RowsPerStrip           0116   0004   00000001  00001130
   00AE        StripByteCounts        0117   0004   00000001  <COUNT0>
   00BA        XResolution            011A   0005   00000001  00000164
   00C6        YResolution            011B   0005   00000001  00000164
   00D2        XPosition              011E   0005   00000001  0000016C
   00DE        YPosition              011F   0005   00000001  0000016C
   00EA        Group4Options          0125   0004   00000001  00000002
   00F6        ResolutionUnit         0128   0003   00000001  00020000
   0102        Software               0131   0002   00000008  00000174
   010E        DateTime               0132   0002   00000014  0000017C
   011A        Artist                 013B   0002   00000008  00000190
   0126        HostComputer           013C   0002   0000000F  00000198
   0132        Next IFD Pointer       00000000

Fields Offsets Point to:

   0136        DocumentName          "LAMap1"
   013E        ImageDescription      "A map of Los Angeles"
   0154        Make                  "Fujitsu"
   015C        Model                 "M3093E"
   0164        X,Y Resolution        00000190 00000001
   016C        X,Y Position          00000000 00000001
   0174        Software              "Xionics"
   017C        DateTime              "1990:10:05 15:00:00"
   0190        Artist                "Joe Pro"
   0198        HostComputer          "Tardis.Isi.Edu"

Image Data (<STRIP0> from above is here 01A8)

   01A8        Compressed Data for single strip, of length <COUNT0> bytes

[end of TIFF file]

NOTE:  Since in this example there is only a single strip, there is only
       one count for StripByteCounts and one offset for StripOffsets.
       Thus, each of these only takes 4 bytes and will fit in the
       Value Offset instead of being pointed to.

5. Conclusions

Bitmapped images transferred within the Internet should be in the following format:

1. The file format should be TIFF-B with multi-page files

supported. Images should be encoded as one TIFF strip

per page.

  1. Images should be compressed using MMR when possible. Images may also be MH or MR compressed or uncompressed. If MH or MR compression is used, scan lines should be "byte-aligned".
  1. For maximum interoperability, image resolutions should either be 600, 400, or 300 dpi; or else be one of the standard Group 3 fax resolutions (98 or 196 dpi vertically and 204 dpi horizontally).

Note that this specification is self contained and an implementation should be possible without recourse to the TIFF references, and that only the specific TIFF documents cited are relevant to this specification. Updates to the TIFF documents do not change this specification.

Existing commercial off-the-shelf products are available which can handle images in the above format. ISI would be delighted to help those interested in assembling a system.

6. Acknowledgments

Many contributions to this work were made by members of the IETF Network Fax Working Group especially by its chairman, Mark Needleman and by Clifford Lynch of the University of California Office of the President, Library Automation. Also, Kiyo Inaba of Ricoh Co. Ltd. made a number of helpful suggestions.

7. References

   [1] Borenstein, N., and N. Freed, "Mechanisms for Specifying and
       Describing the Format of Internet Message Bodies", RFC in

[2] International Telegraph and Telephone Consultative Committee

(CCITT), Red Book, October, 1984.

   [3] Aldus Corp., Microsoft Corp., "Tag Image File Format
       Specification", Revision 5.0, Final, 1988.
   [4] Cygnet Corporation, "The Spirit of TIFF Class F, 1990", available
       from Cygnet Technologies, 2560 9th., Suite 220, Berkeley, CA
       94710, FAX: (415) 540-5835.
   [5] Welch, T., "A Technique for High Performance Data Compression",
       IEEE Computer, Vol. 17, No. 6, Page 8, June 1984.

8. Security Considerations

While security issues are not directly addressed by this document, it is important to note that the file format described in this document is intended for the communications of files between systems and across networks. Thus the same precautions and cares should be applied to these files as would be to any files received from remote and possibly unknown systems.

9. Authors' Addresses

   Alan Katz
   USC Information Sciences Institute
   4676 Admiralty Way #1100
   Marina Del Rey, CA  90292-6695
   Phone: 310-822-1511
   Fax:  310-823-6714
   EMail: Katz@ISI.Edu
   Danny Cohen
   USC Information Sciences Institute
   4676 Admiralty Way #1100
   Marina Del Rey, CA  90292-6695
   Phone: 310-822-1511
   Fax:  310-823-6714
   EMail: Cohen@ISI.Edu