PROJECTS > Image and Video Processing
RTUK SKAAS - KAVTAN (CONCEPT RECOGNITION SYSTEM)
Concepts are general abstractions of objects or ideas in human mind. Concept recognition or high level feature extraction is the the process of enabling the computers to be able to recognize and detect certain concepts automatically. In this respect, textual searh reached a mature state after years of extensive research and became a part of daily life. However, a similar search is still difficult for multimedia data.
Despite this difficulty, concept recognition is an indispensable part of modern multimedia archives. This importance increases for large scale archives, such as RTUK-SKAAS, since human effort can not span large amount of data practically.
In KAVTAN Project, the ability of automatic concept recognition on audiovisual data will be introduced to RTUK-SKAAS System. For this purpose, certain concepts, which are within the interest area of RTUK Monitoring Department, are predetermined. The system that automatically recognizes and tracks these concepts in broadcast media will be developed and integrated to RTUK-SKAAS System.
RTÜK SKAAS Project
SKAAS is a large scale digital media archiving and content management project. The client of the project, Radio and Television Supreme Council of Türkiye, demands monitoring of more than 300 TV channels and 900 radio stations which constitutes all the broadcast content of Turkiye. TÜBİTAK UZAY, with its expertise on audio/image/video processing and pattern recognition, is responsible for the analysis requirements of the project such as:
Query by Example Video
Retrieval of similar visual content is one of the fundamental properties of a multimedia archive. In this way, browsing the archive and finding the similar content are highly facilitated. The key contents of the module are indexing and query. In the indexing part, all the visual content in the archive is indexed in a compact and descriptive way. Then these index values are utilized in the query in a fast manner and similar contents in the archive can be obtained.
Video Shot Boundary Analysis
In the shot boundary analysis, the shots constituting the whole video are found and represented by a keyframe that summarizes the content of the shot. Different low level sources of information, such as color, edge and motion, are utilized to detect hard and gradual shot boundaries. Then a representative frame is selected as the keyframe of the shot. In this manner, the user is presented a brief summary of the video in terms of still images. Hence, browsing the archive without watching the whole video is possible.
Video Text Localization and Recognition.
For the indexing and management of large scale video databases an important tool would be the ovelay text in the digital media. Such information, for example, can be used to index any video database quite efficiently and effectively. Speaker information, location, date/time, score results of sport games, etc. can be queried more thoroughly, since these information can be determined as text data in the digital media. In this work, a system, that localizes and recognizes overlay text containing Turkish and/or English characters in captured television broadcasts, has been developed.
Keyword-spotting
Today, use of multimedia data for personal and institutional purposes is increasing rapidly. As a consequence, fast and efficient archiving of that large amount of data and data retrieving from multimedia archive becomes a problem. Keyword-spotting has a very convenient and effective usage for audio data archiving and audio retrieval applications. Keyword-spotting application provides automatic detection of the locations of any word or word sequence from television and radio broadcast data. System can track any utterance independent from speaker. The utterance to be spot is given as an input to the system in text format and needs no supervised training. Keyword-spotting application provides high performance rates for audio retrieval and represents a reliable system.
Radio Commercial Interval Analysis Application
Tracing the commercial intervals on broadcast radio is very important for many advertisement agencies and commercial regulation commissions. When the number of national and local radio channels is considered, it is very difficult to continuously follow commercial intervals manually. Radio Commercial Interval Analysis Application is developed to trace commercial intervals automatically using the technology of commodity processors. The application utilizes the commercial jingles that are broadcasted at the boundaries of commercial intervals as references to report the positions of commercial periods. Providing one sample for each commercial jingle to the system is sufficient for this application. The application provides a system for reliable, fast and robust detection of commercial intervals and generates commercial interval reports automatically.
GÖRETAS Project
GÖRETAS is a video processing system specialized for commercial detection and recognition in TV broadcast. System automatically detects commercials, classifies ads inside them, and reports the results to the related entities in approximately ten minutes. In addition, the system detects in real-time the sponsor screens that are seen in TV broadcasts randomly outside commercial blocks.
MPEG-7 Compatible Digital Video Management System Design and Implementation (BilMMS) Project
Efficient indexing, classification and retrieval of multimedia data have become a necessity as a result of the increase in digital multimedia content. BilMMS project facilitates the automatic summarization of multimedia clips and the extraction of features defined by the MPEG-7 standard to index and classify multimedia data in a fully MPEG-7 compliant manner.
GAYE Project
GAYE is a face recognition system implemented with the special purpose of surveillance applications. As the major advantage of the proposed system is its robustness to occlusion and illumination changes, it is also particularly suitable for database searches to identify a person.
A novel approach is used for both feature extraction and face comparison. Face recognition method of GAYE achieves great performance results compared to the other methods in the literature.
System can also detect, track, find the best pose and recognize faces from real-time video.