When choosing a company’s DAM solution, there are five key criteria that should be evaluated. The relative importance of any criterion will vary based on the strategic objectives and particular situation of your company and should be adjusted accordingly.
- Scalability of platform
- Versatility of solution (media files supported, specialized applications)
- Accuracy of results
- Speed of indexing/retrieval
- DAM System Type
The first four criteria are self-explanatory. However, the fifth require some additional context. Typically, DAM systems were designed to serve one of two purposes:
- Search Indexing consists of cataloging thumbnails of original media in an indexed database that can be searched using keywords. The source files are left untouched. At present time, most indexing approaches rely on embedded tags manually encoded into files. An accurate and reliable computerized method for visual query of still images and comparison of visual similarity of video shots to “create and embed tags into files” is an open problem with ongoing research
- Asset repositories store the source files in a secure database. Benefits include security, referential integrity, and centralized data management along with full storage management and disaster recovery. DAM systems based on the asset repository model are most suitable for managing rights and access permissions that include intellectual property rights, and structuring global access by customers. The centralization of assets requires significantly higher performance hardware such as high-end UNIX servers, substantial online storage, and very high-speed networks. Capital investment requirements for catalogs are 10x to 50x the cost of catalogs.
Based on the five criteria listed above, we have evaluated a number of DAM products to determine their suitability for the multimedia search and content management. We found many jack-of-all-trades asset management solutions. However, these solutions tend to have crude and rudimentary search functionality. For best-in-class multimedia search capabilities, we suggest you look into Virage for still-images and video, and Truveo or Blinkx for video-only search. Additionally, Microsoft is developing next-generation tools that may alter the competitive landscape in 4-7 years. Below are details on each company.
San Francisco, CA 94105
Tel: (415) 243 9955The Virage video indexing applications offer an array of commercial grade features, plus additional specialized applications (such as surveillance video analysis). The system quickly indexes video and still images, and encodes it for real-time searchability. Virage is a wholly owned subsidiary of Autonomy, Inc, a UK-based company with a $2B market cap.
Cost of the Virage software is between $200k-$500k if used internally. If the system is “monetized” for use on a website, the company would charge $1M
An AOL Company
333 Bush Street, 23rd Floor
San Francisco, CA 94104
Phone: 415-844-9000Truveo powers many of the main video search sites including AOL, Microsoft, Search.com, Excite, and Infospace. In 2006, it released the Truveo Developer site, which provides a variety of APIs for 3rd party sites.
blinkx Inc. (London AIM:BLNX.L)
One Market, 19th Floor, Spear Tower
San Francisco, CA 94105, USA
Phone: 415 848 2986
Blinkx’s accuracy in searching video comes from analyzing the closed captions and using speech recognition (English, German and Spanish) to generate transcripts. Image analysis is also done, though on a lesser scale. Their website boasts of 14 million hours of searchable videos.
Microsoft Live LabsSources at the National Security Agency report that Microsoft’s blue-sky research is head and shoulders above other companies. Two products under development illustrate these advances:
- Photosynth is a tool that takes images of related topics and seamlessly reconstructs them into a navigable universe
- Seadragon is a technique that removes the computational performance barriers of image search and display
These technologies do not directly solve the problem of multimedia search – since they require a collection of related photos. However, as these tools are developed, they will be instrumental in changing the way we search.
Digital Video Multimedia Lab
Department of Electrical Engineering
1312 S.W. Mudd
500 West 120th Street, New York, NY 10027
Phone: (212) 854-3105Academic research labs such as that at Columbia are developing new algorithms for content analysis. The Digital Video Multimedia Lab is sponsored by AT&T and the National Security Agency. Research focuses on five areas:
- Multimedia indexing and management
- Feature extraction and object/text recognition
- Pervasive media
- Authentication and watermarking
- Multimedia standard, testbed, and evaluation
IBM Multimedia Analysis and Retrieval System
Intelligent Information Management Dept.
IBM T. J. Watson Research Center
19 Skyline Drive
Hawthorne, NY 10532 USAIMARS employs a multi-modal machine learning algorithm. In its demonstration form, it supports 10-20 semantic concepts over a mere few hundred hours of broadcast news clips. A fully functioning version requires a semantic ontology of 1,000 concepts – just to cover broadcast news. Although it is still far from commercial viability, the techniques employed appear very promising, given additional development.
Quick Square LLC
3510 Snouffer Road Suite 200
Columbus, OH 43235
Phone: 866-849-5844Enterprise-grade multimedia management and repository, with keyword-based search and categorization
700 King Farm Boulevard, Suite 400
Rockville, MD 20850
Phone: 301.548.4000Enterprise-grade multimedia management and repository, with keyword-based search and categorization. Typical license costs $100k – $200k, with customization available.
APPENDIX: Technical Approaches to Video Indexing
Recent developments in indexing videos have used optical recognition software to scan video files for closed captioning. Speech recognition software employing neural networks and machine learning using “hidden Markov models” can index video files that do not have closed captioning.
An accurate and reliable computerized method for visual query of still images and comparison of visual similarity of video shots is an open problem with ongoing research at academic universities and private firms. Video contains an enormous amount of visual information with 25 frames every second, high VGA resolution, and a high color depth. A popular method to simplify the data analyzes key frames as a representative sample of a video scene rather than keeping thousands of frames. When there is motion or large variance in the sequence, key frames may not capture complete visual information.
Researchers are proposing an algorithm that uses motion estimation to generate a series of key frames based on detection of significant camera motion. Producing even a few key frames of a scene presents a serious indexing problem along with redundancy clutter.
The standard approaches are the “query by example” and “query by keyword” models. Research is now exploring “probabilistic semantic indexing” to accommodate concepts in addition to key words.