Project motivation
Initial situation
10 years of experience in audio digitization
In the year of 1999 the Austrian Mediathek started to think about digitization. At that time some institutions have already been digitizing books and other printed originals. As the Austrian Mediathek is an archive dealing with audio and video content, the existing tools focussing on printed media were not usable. Video was far away from reasonable technical solutions. So it was audio that was to be digitized first at the Austrian Mediathek. Together with Christophe Kummer, the later founder of NOA, tools for digitizing audio recordings were developed and are still in use. Although there was no trained developer inhouse, during the last ten years a formidable knowledge about digitization and longterm archiving grew at the Austrian Mediathek.
Beside digitizing more than 60,000 of the own recordings, the institution managed to achieve contracts with other organisations to digitize their audio recordings too. Digitizing for the needs of external archives was an additional skill. It required more precise documentation and more comprehensible workflows.
Video stock at the Austrian Mediathek
From the actual existing 410,000 audio carriers there are still many to digitize. But it showed that the additional income from the external digitization projects was necessary to expand and support the needed equipment. There is still plenty to do and there are also 35,000 video carriers waiting for the migration. The video formats seem to vanish more quickly. Their lifetimes are shorter. Their short period of existence reduces the chance to find used equipment on the market, when manufacturers stopped building them. And at last, unlike some oldfashioned audio formats as 78s or vinyl, because of their complexity and in many cases undocumented and often proprietary hardware it is almost impossible to rebuild video players.
Despite the knowledge of this urgent need to start digitizing video the Austrian Mediathek hesitated. There was still no solution for a reasonable archive format not to say for a complete system to do the job of digitizing.
The project "Österreich am Wort"
In September 2009 the Austrian Mediathek started a project called "Österreich am Wort". Aim of this project is to digitize about 10,000 recordings and publish at least 5,000 of them on the web until the August of 2012. This includes 1,000 video recordings. Therefore there was a necessity to find and start using a solution for video digitization. With the advantage of long experience in audio digitization but a tight time schedule the planning for video digitization began.
Basic needs
Talking about digitizing for an archive means not only to think about a high quality conversion of video and audio information into the digital, file based world. It also requires a sense for the needs of a longterm archive. Some decisions, for example which archive format to select or whether to read out actual metadata of tools used in the conversion chain, seem to be irrelevant for the current need of just a digital viewing copy. But they will be essential for later use.
Archive format
Unlike audio digitization where the community decided to use the Broadcast Wave Format (BWF) with 96 kHz / 24 Bit resolution. See also: Publications of the IASA, there is still no widely accepted archive format for video. Here is a list of requirements, a codec used for long-term archiving of video must fulfill:
Uncompressed/lossless
The most important requirement for the archive format is that it is absolutely lossless. Consequentially, as the content is expected to last forever, it must undergo endless instances of conversion into future formats. Therefore each loss in quality, even if it is minimal, would - at the end of the migration chain - lead to a total loss of content.
Not proprietary
The codec must not be proprietary. It must be capable of being used by any programmer for implementation in any program that is intended to be dealt with it. Otherwise the content of the archive depends on the will and faith of the one-, and in many cases only software-manufacturer. See also: OAIS model, chapter 4.2.1.3.2 Representing Network
Hardware independency
Playback of the archive's content also must not be depending on proprietary hardware. This would limit the use of programs to those that can deal with the dedicated hardware, which in many cases again would be proprietary software. Only hardware that could also be build by other than the actual manufacturer should be used for extracting and interpreting the archive files.
Handleable data amount
Calculations have shown that video digitization produces a huge amount of data. Although the prices for storage decrease steadily, the costs are still very high. We are talking about a vast amount of terrabytes here, so calculating in these dimensions, the chance to cut them in half becomes an important argument.
SD/HD/2K resolution
The actual video stock in the archive of the Austrian Mediathek is not exceeding SD in resolution. But with HDV recordings entering the archive as part of some collections and HD being adopted even in consumer cameras and therefore soon to be expected as input for the archive, the resolution of the archive file format to choose must not be limited by the SD-resolution standard. High quality scans of film stocks, even if it is just 8mm film, will be done in 2K.
Aspect ratio
Of course the old fashioned and well represented 4:3 recordings as well as the newer and meanwhile dominating 16:9 material have to be covered in the archive format.
Colorspace
Different video formats deal with different colorspaces. Because converting of colorspace implicates some inevitable losses, it should not be altered. Therefore the file codec must be able to map all eventually occuring colorspaces, including their subsampling options.
Known alternatives for an archive codec
When you strike all lossy and proprietary codecs off the list of known available video formats, only two well known alternatives seem to be left:
* UncompressedUncompressed formats require a huge amount of storage space, which -depending on budget limits- eventually means limiting the overall digitizing efforts.
* JPEG 2000This codec wrapped in an MXF container is considered to be an excellent solution for a video archive format in many discussion circles and was, therefore, our favorite solution until we made tests to deal with those files. Many of the few existing applications, which claim to deal with this format, only are able to decode the file. Others cannot handle the MXF container. It also seems difficult to play back a lossless JPEG 2000 file properly without dedicated hardware, because the algorithm of the codec is very resource demanding. This dependency on hardware makes JPEG 2000 (a primarily open format) partially proprietary.
Both known alternatives are at least viable possibilities for storing video content without facing a loss in quality through successive migrations. But none of them fulfills all of the requirements as stated in this text. Although they seem to represent a serious solution for a long sought reliable format for long term video archiving, many technicians do not seem to be completely satisfied with them.
The research at the Austrian Mediathek also included the area of Open Source Software and discovered a number of additional candidates for a possible archive format. The most interesting codec was "FFmpeg's lossless codec v1" - short: FFV1.
* FFV1The fact that almost all programs in the Free Software community dealing with audio and video are relying on the libavcodec library, which is part of the ffmpeg project, makes all the codecs and containers, which are implemented in that library, well distributed among available Free Software tools. This library is so open, free and well-known that some commercial vendors have even decided to implement it in their products. For example: Rhozet Carbon Coder, NOA Audio Solutions, and others. Thanks to wrapper applications for several different operating systems (perion for Mac, ffdshow-tryouts for PC), the library's multimedia codecs can be used by any application that supports the use of external codecs. On the Microsoft Windows platform, for example, one can use "ffdshow-tryouts", which makes the libavcodec available as VfW (Video for Windows) and DirectShow codec. After a successful installation, all applications, which are able to deal with system codecs - whether they are Free Software or proprietary - now also can handle all formats and codecs inside this library.
One of these codecs, which is part of the ffmpeg library, is the so-called "FFV1". It is a codec with a mathematically lossless compression. It decreases the amount of data down to almost 30 %, which is comparable to JPEG 2000. It is capable of dealing with all current color spaces like YUV, YV12 and RGB including subsampling (4.4.4, 4.2.2 and so on). It can even handle arbitrary resolutions. We tried PAL 4:3 and 16:9 as well as HD 1980 x 1080.
But we made no tests with NTSC. As the Austrian Mediathek is located in Europe and is a national AV archive, we have only video with the PAL standard in our collections. Tests with NTSC would be part of future investigations. The great advantage of the FFV1 codec is its comparatively low demand on processing power. Using this codec we achieved proper recordings without adding or dropping frames using a PC with currently off-the-shelf-regular hardware. Even the playback performed fluently.
Managing the workflow
To deal with different formats and several ingests in term and still keep track of the process, it becomes important to manage the workflows.
Automatisms
To spare time and/or get the work done as accurately as possible, the workflows need to be split into those parts that have to be done manually and those which are done by an automatism. Performing the automatic parts in the backround enables to go on with the next recording instead of waiting until the operator's client workstation has finished its tasks.
Monitoring
Splitting the workflows into several tasks which eventually can't be performed at once asks for a concept to access and view the status of each recording.
Error handling
If some problems or mistakes occur with a recording, there must be a procedure to stop/pause the workflow and reset tasks for another try with changed parameters. Having several parallel workflows running such an interference mustn't affect the other workflows.
Technical metadata
Documenting the whole process of digitization including metadata of all tools used and their configuration is important for later research.
Collecting data
Every task should be protocolled. Every used tool whether it is hardware or software is not only to be enumerated. Detailed information about its brand, version and configuration becomes vital when the quality of the recording gets called into question in later times.
XML standard
For easier interpretation of the collected data they should be written in a comprehensible way. Using an existing standard makes it even more transparent.
Getting trustworthy data
There is a difference between configuration documented manually and those read by machines. In the first case it is information that a person thinks is correct. In the latter case even small changes in configuration which are usually unnoticed become identified and protocolled.
Need to economize digital storage space
The amount of storage space needed for high quality archive video files is so high that even now, when prices for disks dropped very far compared to five years before, it is essential for budgets whether the space needed can be reduced to a third or not. Lossless compression therefore is a vivid prospect in the field of digital video archiving.
Limited budget for project
Only a small portion of the budget could be spent on video digitization at the Austrian Mediathek. There was only a limited funding for the project "Österreich am Wort". Therefore, the system to be installed for the video digitization had to be cost-effective.