Theory and applications of
Pattern Recognition and
Computer vision
We work on research in both viewpoints of theory and practice.
Theory: considering deeply the physical natures of an addressed problem
Practice: pursing concrete problems with higher social needs.
Theory
Pattern Recognition
- Theory construction of the subspace based methods for classification
- Multiple view image processing
- Fusion of "appearance based method" and "geometry based method"
What is "Subspace Method?"
Subspace Method is one of very importance and effective methods for pattern recognition and computer vision. Even now, various new types of the methods based on the Subspace Method, such as "Mutual Subspace Method", have been still proposed, although Subspace Methods have been practically used for over 30years.
Subspace methods are used for
- Character recognition
- Face recognition
- Object recognition
- Voice recognition
- Text classification
- Data mining
If you are interested in the methods related to "Subspace Method", please visit the homepage of Workshop on Subspace Methods. There are so many latest theses about "subspace method" on the proceedings here.
Workshop on Subspace Methods
- Subspace2006(MIRU2006 workshop)
- Subspace2007(ACCV2007 workshop)
- Subspace2008(MIRU2008 workshop)
- Subspace2009(ICCV2009 workshop)
Fusion of Statistic theory and geometric theory in pattern recognition
- Fusion of Subspace method and Factorization method
Learning
- Method and system for learning with CG images or image sequence from TV
Extracting features from image patterns
- Edge (Separability filter, etc)
- Position invariant features, such as HLAC, CHLAC
- Nonlinear feature extraction
Object recognition
- Face and 3D object recognition using multiple view image sequence
- Biometrics
- Robot vision based on image sequences
Human sensing
- Human detection and tracing
- Motion recognition
- Gazed area detection
- Image based human interface
Understanding situation
- Distributed multiple camera system
- Video surveillance
- Unusual detection
- Implementation of various recognition method on FS (Field Server)
Hardware for image proccessing
- Image recognition using image processing board "Visconti"
Advanced themes
- Fusion of notion understanding and image recognition
Mutual Subspace Method (MSM) is mathematically same as Canonical Correlation Analysis (CCA). Thus, a notion space may be theoretically connected to a discrimination space generated by Mutual Subspace Method. - Fusion with bioinformatics
Our Constrained Mutual Subspace Method(CMSM), Kernel constrained mutual subspace method(KCMSM)), etc, are not only used for image recognition but also various classification problems. I have a plan of applying our methods to an interesting problem related to bioinformatics.