Extracting Recording Date & Time from AVI film clips.

Last updated: 11 Jan 2010

Introduction

There seems to be little information around on the web over how to extract the recording-date and time from film clips. As digital film and photos allow people to record larger quantities of visual data, we also become dependent on having efficient ways of archiving this material. One important step in this process is to be able to identify the date and time a particular photo or film clip was taken. This page attempts to give some help on doing this in the Java language.

Relevant links

• Drew Noakes has made a great Java library, ExifExtractor for extracting the Exif data out of jpeg files. The Exif data contains the date and time a photo was taken, as well as loads of other interesting things, such as exposure settings, compression, and so on.

• The open-source Kino project has a lot of C code for processing film clips.
  
  
• MediaInfo is an open source project in C++ that reads meta-data from film clips.

Read Eden's latest article: Empowering Teenage Girls in Africa

The eden-avi-lib library

This page presents Eden Foundation's library for extracting the dates out of AVI film clips. It is a small Java library, supporting files with DV encoding and MJPEG encoding. Clips encoded in MJPEG format tend to store their date stamps in different ways, depending on the manufacturer, and this code is capable of extracting the recording times stored by FujiFilm and Samsung cameras. It may work for other brands as well, but has not been tested on them.

Click here to download the library (Version 0.1.4, published 2010-01-11)

You are free to use this library, including the source code, in any way you see fit.

Command-line useage:

java -jar eden-avi-lib-0.1.4.jar /my_path/my_film.avi

The date stamp formats

Here is a technical discussion over how eden-avi-lib extracts various dates:

DV-AVI

DV-AVI files can be encoded in two subformats: ivas-dvsd (Type 1) and vids-dvsd (Type 2). The difference between the two types is that the Type 1 encoding lets the video and audio share the same track, whilst for Type 2, the video and audio data are on separate tracks in the AVI file. Eden-avi-lib only supports DV-AVI files using the Type 2 encoding. It searches for the first frame in the file, and reads the date stamp encoded there.

MJPEG-AVI

MJPEG-AVI files store their date stamp in a chunk in the header of the file, rather than in the video track itself. However, there are several different ways of storing this stamp, depending on the manufacturer of the camera used to record the film clip. Eden-avi-lib is currently capable of reading three such chunks:

• The LIST/HDRL IDIT chunk (used by Samsung and Canon cameras). Once identified in the AVI file, it contains a single string containing the recording date and time. The format may vary depending on camera manufacturer:

  Manufacturer	Sample IDIT string
  Samsung	2005:08:17 11:42:43
  Canon	THU OCT 26 16:46:04 2006
  Fujifilm	Mon Mar  3 09:44:56 2008

  
  Note that in the case where the month name is represented as text, the day number will not necessarily be zero-padded when it is under 10 (look at Fujifilm above).
  
  
• The LIST/strd chunk (used by Fujifilm cameras). This chunk contains mainly binary data that is hard to understand. It always starts with the four characters "AVIF". On the 99th byte of the data chunk, a number of null-terminated character strings appear. The first is "FUJIFILM", followed by an extra null-terminator character. After this comes the camera's model name (usually FinePix****), sometimes but not always followed by a surplus null-terminator character. Following this comes the recording date and time. After the recording date and time comes a lot of binary data that again is hard to figure out.
  
  There are two peculiar details with this chunk:
  1. The null-terminated camera model string is sometimes followed by a surplus null character, sometimes not. Here is a table over the analysis of film clips taken with various FujiFilm camera models:
     
     Basic format: FUJIFILM  FinePixXXXX  200X:XX:XX XX:XX:XX
     Initial offset: 98 bytes.
     
     

     Model	Nz1	Nz2	Offset of Date
     4700Z	2	2	122
     30i	2	2	120
     2600Zoom	2	1	124
     F410	2	2	124
     A201	2	1	120
     50i	2	2	120
     4500	2	1	120
     2650	2	2	124

     
     Nz1 and Nz2 show the number of null characters following "FUJIFILM" keyword and the model name.
     Offset of date shows where the date string itself starts within the strd chunk.
     
     
  2. The "strd" chunk is actually one byte longer (a null character) than what the four bytes following the "strd" key indicate. This can make parsing the avi header somewhat trickier, if one needs the data following this chunk.

• The LIST/ncdt chunk (used by Nikon P6000 and possibly other Nikon cameras). It starts with the four characters "ncvr". We find the date here at an offset of 222 bytes within the chunk, in a format like "2008:10:06 10:55:48". Thus parsing is simple.

Doing it for QuickTime movies

QuickTime movies (mov files), at least those recorded by Konika Minolta cameras, store the recording date in the "mvhd" atom. To access the different atoms, Chris Adamson has written an article and a piece of code. The format of the "mvhd" atom is described on Apple's developer site. Basically, we want to extract the 5th to 8th bytes of the atom, and then convert those four bytes to a Calendar object. Here is a method that performs such a conversion:

 private Calendar bytesToCal(byte[] buffer) {
   int i = 0;
   int pos = 0;
   i += unsignedByteToInt(buffer[pos++]) << 24;
   i += unsignedByteToInt(buffer[pos++]) << 16;
   i += unsignedByteToInt(buffer[pos++]) << 8;
   i += unsignedByteToInt(buffer[pos++]) << 0; 
   long l = unsignedIntToLong(i);
   
   long days = l / 86400;
   long time = l % 86400;
   long hours = time / (60 * 60);
   long minutes = (time % (60 * 60)) / 60;
   long seconds = time % 60;
   
   GregorianCalendar cal = new GregorianCalendar(1904, 0, 1);
   cal.add(Calendar.DAY_OF_MONTH, (int)days);
   cal.set(Calendar.HOUR_OF_DAY, (int)hours);
   cal.set(Calendar.MINUTE, (int)minutes);
   cal.set(Calendar.SECOND, (int)seconds);
   
   return cal;
 }
 private static int unsignedByteToInt(byte b) {
   return b & 0xFF;
 }
 private static long unsignedIntToLong(int i) {
   return i & 0xffffffffL;
 }

Special Thanks

Special thanks go to Johan Stäck, who solved the puzzle of extracting the recording date out of DV-AVI frames, a task that requires quite some binary acrobatics. His code was written in Visual Basic, and I ported it to Java. Unfortunately, Johan's page is down now (lightweightvideo.com).

Contact

For feedback, please contact Josef Garvi on "josef at eden-foundation dot org". If anyone has tips on how to extract these time stamps from other types of film clips, please let us know!

Eden Foundation Home Page