Pattern Recognition in Multimedia Data
»Multimedia Pattern Recognition« involves the research and development of pattern recognition processes, tools and software solutions for voice, image, audio and video data as well as documents. The surge in digital data has increased demand for automatic analysis and recognition methods. This type of evaluation will frequently need to be undertaken in real time such as for the automatic recognition of traffic signs or the generation of subtitles during live broadcasts. Evaluating such enormous amounts of data manually would also be prohibitively expensive.
Methodology: Data-driven Procedures and Learning Procedures
Two main issues need to be taken into account when researching and developing pattern recognition technologies: First of all it is usual for statistical classifiers to be used for recognition purposes before being trained using extensively annotated data pools. These types of data-driven methods lead to the robust recognition technologies which make the use of learning procedures possible. Secondly, pattern recognition technology is selected and developed to ensure it is suitable for real-life scenarios. Specific application scenarios present particular challenges to research and development teams working in this area.
Stable Recognition Technology for Long Runtimes
Statistical classifiers such as deep neural networks, support vector machines and hidden Markov models are applied to the transformed input data as part of the actual recognition process. As the recognition technology is being used in a productive environment it is important that it is particularly stable in terms of its runtime behavior and maintainability.