TIFF Image Creation

Written by Paul Bourke
August 1998


The following demonstrates how to create 24 bit colour RGB TIFF (Tagged Image FIle Format) files. That is, how to create images from your own software that can be then opened and manipulated with image handling software, for example: GIMP, PhotoShop, etc. Given this aim, this document illustrates the "minimal" requirements necessary to create a TIFF file, it does not provide enough information for writing a TIFF file reader. For more information on the full TIFF specification the following postscript and pdf files describe the TIFF version 6.

tiff.ps.gz       tiff.pdf       summary.pdf

The basic structure of a TIFF file is as follows:

The first 8 bytes forms the header. The first two bytes of which is either "II" for little endian byte ordering or "MM" for big endian byte ordering. In what follows we'll be assuming big endian ordering. Note: any true TIFF reading software is supposed to be handle both types. The next two bytes of the header should be 0 and 42dec (2ahex). The remaining 4 bytes of the header is the offset from the start of the file to the first "Image File Directory" (IFD), this normally follows the image data it applies to. In the example below there is only one image and one IFD. /

An IFD consists of two bytes indicating the number of entries followed by the entries themselves. The IFD is terminated with 4 byte offset to the next IFD or 0 if there are none. A TIFF file must contain at least one IFD!

Each IFD entry consists of 12 bytes. The first two bytes identifies the tag type (as in Tagged Image File Format). The next two bytes are the field type (byte, ASCII, short int, long int, ...). The next four bytes indicate the number of values. The last four bytes is either the value itself or an offset to the values.

Considering the first IFD entry from the example gievn below:
      0100 0003 0000 0001 0064 0000
      |    |    |         |
tag --+    |    |         |
short int -+    |         |
one value ------+         |
value of 100 -------------+
Example

The following is an example using the TIFF file shown on the right, namely a black image with a single white pixel at the top left and the bottom right position. The image is 100 pixels wide by 200 pixels high.

A hex dump is given below along with matching pointers and locations marked in matching colours, these colours further match the appropriate parts of the source code gievn later. The tags are underlined.

   4d4d 002a 0000 ea68 ffff ff00 0000 0000
   0000 0000 0000 0000 0000 0000 0000 0000

   .... ....  black 0's deleted  .... ....

   0000 0000 00ff ffff 000e 0100 0003 0000
   0001 0064 0000 0101 0003 0000 0001 00c8
   0000 0102 0003 0000 0003 0000 eb16 0103
   0003 0000 0001 0001 0000 0106 0003 0000
   0001 0002 0000 0111 0004 0000 0001 0000
   0008 0112 0003 0000 0001 0001 0000 0115
   0003 0000 0001 0003 0000 0116 0003 0000
   0001 00c8 0000 0117 0004 0000 0001 0000
   ea60 0118 0003 0000 0003 0000 eb1c 0119
   0003 0000 0003 0000 eb22 011c 0003 0000
   0001 0001 0000 0153 0003 0000 0003 0000
   eb28 0000 0000 0008 0008 0008 0000 0000
   0000 00ff 00ff 00ff 0001 0001 0001
Example image

The above example uses 14dec (000eihex) directory entries.
0100 - Image width
0101 - Image height
0102 - Bits per sample (8)
0103 - Compression method (1 = uncompressed)
0106 - Photometric Interpretation (2 = RGB)
0111 - Strip Offsets
0112 - Orientation (1 = 0 top, 0 left hand side) 0115 - Samples per pixel (1)
0116 - Rows per strip (200 = image height)
0117 - Strip Byte Counts (60000 = 100 x 200 x 3)
0118 - Minimum sample value (0,0,0)
0119 - Maximum sample value (255,255,255)
011c - Planar configuration (1 = single image plane)
0153 - Sample format

Source code example

The following is the guts of a C program to create a TIFF file of width nx, height ny. Each pixel is made up of 3 bytes, one byte for each of Red, Green, Blue. Each colour component ranges from 0 (black) to 255 (white).

   /* Write the header */
   WriteHexString(fptr,"4d4d002a");    /* Big endian & TIFF identifier */
   offset = nx * ny * 3 + 8;
   putc((offset & 0xff000000) / 16777216,fptr);
   putc((offset & 0x00ff0000) / 65536,fptr);
   putc((offset & 0x0000ff00) / 256,fptr);
   putc((offset & 0x000000ff),fptr);

   /* Write the binary data */
   for (j=0;j<ny;j++) {
      for (i=0;i<nx;i++) {
 ... calculate the RGB value between 0 and 255 ...
         fputc(red,fptr);
         fputc(green,fptr);
         fputc(blue,fptr);
      }
   }

   /* Write the footer */
   WriteHexString(fptr,"000e");  /* The number of directory entries (14) */

   /* Width tag, short int */
   WriteHexString(fptr,"0100000300000001");
   fputc((nx & 0xff00) / 256,fptr);    /* Image width */
   fputc((nx & 0x00ff),fptr);
   WriteHexString(fptr,"0000");

   /* Height tag, short int */
   WriteHexString(fptr,"0101000300000001");
   fputc((ny & 0xff00) / 256,fptr);    /* Image height */
   fputc((ny & 0x00ff),fptr);
   WriteHexString(fptr,"0000");

   /* Bits per sample tag, short int */
   WriteHexString(fptr,"0102000300000003");
   offset = nx * ny * 3 + 182;
   putc((offset & 0xff000000) / 16777216,fptr);
   putc((offset & 0x00ff0000) / 65536,fptr);
   putc((offset & 0x0000ff00) / 256,fptr);
   putc((offset & 0x000000ff),fptr);

   /* Compression flag, short int */
   WriteHexString(fptr,"010300030000000100010000");

   /* Photometric interpolation tag, short int */
   WriteHexString(fptr,"010600030000000100020000");

   /* Strip offset tag, long int */
   WriteHexString(fptr,"011100040000000100000008");

   /* Orientation flag, short int */
   WriteHexString(fptr,"011200030000000100010000");

   /* Sample per pixel tag, short int */
   WriteHexString(fptr,"011500030000000100030000");

   /* Rows per strip tag, short int */
   WriteHexString(fptr,"0116000300000001");
   fputc((ny & 0xff00) / 256,fptr);
   fputc((ny & 0x00ff),fptr);
   WriteHexString(fptr,"0000");

   /* Strip byte count flag, long int */
   WriteHexString(fptr,"0117000400000001");
   offset = nx * ny * 3;
   putc((offset & 0xff000000) / 16777216,fptr);
   putc((offset & 0x00ff0000) / 65536,fptr);
   putc((offset & 0x0000ff00) / 256,fptr);
   putc((offset & 0x000000ff),fptr);

   /* Minimum sample value flag, short int */
   WriteHexString(fptr,"0118000300000003");
   offset = nx * ny * 3 + 188;
   putc((offset & 0xff000000) / 16777216,fptr);
   putc((offset & 0x00ff0000) / 65536,fptr);
   putc((offset & 0x0000ff00) / 256,fptr);
   putc((offset & 0x000000ff),fptr);

   /* Maximum sample value tag, short int */
   WriteHexString(fptr,"0119000300000003");
   offset = nx * ny * 3 + 194;
   putc((offset & 0xff000000) / 16777216,fptr);
   putc((offset & 0x00ff0000) / 65536,fptr);
   putc((offset & 0x0000ff00) / 256,fptr);
   putc((offset & 0x000000ff),fptr);

   /* Planar configuration tag, short int */
   WriteHexString(fptr,"011c00030000000100010000");

   /* Sample format tag, short int */
   WriteHexString(fptr,"0153000300000003");
   offset = nx * ny * 3 + 200;
   putc((offset & 0xff000000) / 16777216,fptr);
   putc((offset & 0x00ff0000) / 65536,fptr);
   putc((offset & 0x0000ff00) / 256,fptr);
   putc((offset & 0x000000ff),fptr);

   /* End of the directory entry */
   WriteHexString(fptr,"00000000");

   /* Bits for each colour channel */
   WriteHexString(fptr,"000800080008");

   /* Minimum value for each component */
   WriteHexString(fptr,"000000000000");

   /* Maximum value per channel */
   WriteHexString(fptr,"00ff00ff00ff");

   /* Samples per pixel for each channel */
   WriteHexString(fptr,"000100010001");

The WriteHexString() function is given below.

void BM_WriteHexString(FILE *fptr,char *s)
{
   unsigned int i,c;
   char hex[3];

   for (i=0;i<strlen(s);i+=2) {
      hex[0] = s[i];
      hex[1] = s[i+1];
      hex[2] = '\0';
      sscanf(hex,"%X",&c);
      putc(c,fptr);
   }
}


libtiff

While the above creates a simple TIFF file, the ultimate source/API for dealing with TIFF is the library "libtiff". This is particularly so if you wish to read TIFF files which is a non-trivial process if you wish to cover all possibilities. For example, if one wanted to determine the width and height one would not normally search for the width and height tags but rather employ something like this (courtesy Mehmet Vahit Kapar).

#include <stdio.h>
#include <stdlib.h>
#include "tiffio.h"

main(int argc,char **argv)
{
   TIFF* tif = TIFFOpen(argv[1],"r");
   if (tif) {
      uint32 w, h;
      size_t npixels;
      uint32* raster;

      TIFFGetField(tif, TIFFTAG_IMAGEWIDTH, &w);
      TIFFGetField(tif, TIFFTAG_IMAGELENGTH, &h);

      TIFFClose(tif);
   }
   exit(0);
}


Contribution by Chris Rorden

Notes:

  • TIFF RGB images should always have a contiguous (RGBRGBRGB ...) not planar (RRR..RGGG..GBBB..B) format. While the header allows you to specify planar, it is not well supported.

  • For 16-bit data, be aware that most TIFF viewers ignore the maximum and minimum values and assume your data is scaled from 0 to 65535 (You can test this with my example).

  • For 16-bit data, it is important that the endian is set correctly.

  • For data with only one component (e.g. grayscale or black and white) you encode min, max, bits per sample directly in the tag, for RGB data the tag provides a pointer to the address for those values.

The following example are an adaption of the above with the following additional features.

  • Works for grayscale and RGB images.

  • The TIFF writer supports 16-bit images (normalizing values to 65535, getting the endian correct)

  • Supports de-planarize RGB images.
Example that converts popular JPEGs (uses NanoJPEG library)

Example that copies an arcane lossless JPEG format that is popular with medical imaging to TIFF (my own library).