Pattern Recognition Society of Finland
The research area called combinatorial pattern matching has its historical roots in the simple questions of how one can find key-word occurrences in a text. Nowadays the area has developed into a full-fledged special field of algorithmics and data structures with numerous important applications in different fields ranging from information retrieval and data compression to biological sequence analysis in bioinformatics. The talk will give a concise survey of the main algorithmic techniques used in the field as well as describe some applications in bioinformatics (DNA sequencing, gene regulation).
Neuroinformatics combines neuroscience and informatics research to develop and apply advanced tools and approaches essential for a major advancement in understanding the brain. Three dimensional (3-D) brain imaging is extremely central to neuroinformatics, because it is the only technique to investigate the brain structure and function directly in living humans at the system level. Advanced pattern recognition methods are absolutely essential to extract relevant quantitative information from images. Pattern recognition with these data is a unique challenge due to enormous variation in brain structure and function across individuals and the complexity and amount of the information in images. In this presentation, I will give an overview on pattern recognition research and applications in brain imaging and neuoroinformatics in general. I will highlight a few recent important contributions by our group to these fields.
Gaze provides a natural source of information in many information retrieval tasks. In particular, eye movements convey implicit information on the interests of the user, which can be utilized to create more natural search interfaces. We study the use of eye movements in both text and image retrieval settings. Typical text retrieval systems work by matching a query to the set of possible documents. In this setting, eye movements can be used to infer an implicit query based on the way the user is reading documents while performing the search, leading to improved retrieval accuracy. For content-based image retrieval there is even greater need for implicit information sources, since creating good queries is much more challenging. Consequently, a typical CBIR system uses relevance feedback to iteratively refine the results. Explicit relevance feedback is laborious, but can be replaced with implicit feedback inferred from eye movements.
Pattern recognition is an important part of user interface (UI) implementations for mobile devices. They will enable UI flexibility with limited hardware resources. Nokia Research Center has been working on different pattern recognition algorithms to enable new ways to interact with mobile devices and services, for example, using voice, cameras and tactility.
Communicating with mobile devices has become an unavoidable part of our daily life. Unfortunately, the current user interface designs are mostly taken directly from desktop computers. This has resulted in devices that are sometimes hard to use. Since more processing power and new sensing technologies are already available, there is a possibility to develop systems to communicate through different modalities. This talk introduces some new computer vision approaches, including head tracking, object motion analysis and device ego-motion estimation, to allow efficient interaction with mobile devices.
Many central tasks in human-computer interaction like speech recognition, information retrieval and machine translation typically require a large vocabulary. There are several languages (e.g. Finnish, Estonian and Turkish) in which the construction of large vocabulary is particularly difficult due to the high rate of inflection, compounding and concatenative suffixes. In practice, more efficient representation units than words are needed. It is also important to develop models that can learn the suitable sub-word units in a completely data-driven manner and can be easily ported to various morphologically complex languages. In this talk I will present new machine learning based models for state-of-the-art large vocabulary speech recognition and information retrieval and show results of recent multilingual evaluations.
The task is to create a 3D model of a scene using a single standard hand-held camera and a laptop. The model is built at the same time as the user is waving the camera, and the 3D reconstruction is shown on the screen. Our approach for the task is to use Extended Kalman Filter -based visual Simultaneous Localization And Mapping to track the camera 3D-pose, and multi-view stereo to create reconstructions. We introduce the system and on video we show how it works on-line.
Organizers and page maintainers: Erno Mäkinen (etm@cs.uta.fi) and Yulia Gizatdinova (yulia.gizatdinova@cs.uta.fi)