• Home
• News
• Organized events
• Location

• Staff
• Students
• Technicians
• Visitors
• Former members

• Databases
• PhD Theses
• Projects
• Publications
• Software

• Access

Pattern Recognition

Pattern Recognition can be defined as the classification of input signals in different classes, depending on their characteristics. For that, it is necessary to separate those really significant properties from those that are irrelevant details for the classification. Applications of these techniques are numerous: handwritten text and music recognition, teledetection, speech recognition, fingerprint recognition, biomedical signal analysis, etc.

There are two main approaches to Pattern Recognition:

• geometric or statistical methods, in which the signal is described in terms of a set of characteristics (number of curves, number of holes, etc);
• structural or syntactic methods, in which the signal is described in terms of the relationship between its components.

The main worklines of our group are:

• Fast nearest neighbor (NN) search algorithms for metric spaces

  This classifier assigns a sample the class of its nearest neighbor. We have algorithms for finding NN that do not require a vector representation of objects. They are specially suitable for tasks where the distance between two objects is computationally demanding, e.g. the edit distance.

• Classification rules based on neighborhood

  The NN rule and the k-NN rule define a neighborhood around the sample. We have developed alternative neighborhood definitions, like the k-NSN rule, that uses the candidates to NN in a fast search algorithm.

• Combination of classifiers

  There are many ways of combining classifiers for obtaining better results, such as fusion of outputs, using different training sets or different feature sets, or ways of training weights for classifier output combination.

• Application of pattern recognition techniques to:
  • handwritten character and music recognition
  • classification of marble textures
  • music style, genre, and author classification
  • recognition of music melodies
  • robot vision and grasping

Associations and other PR groups:

• International Association for Pattern Recognition (IAPR) www.iapr.org
• Asociación Española de Reconocimiento de Formas e Inteligencia Artificial, AERFAI (spanish association for pattern recognition and artificial intelligence) aerfai.org/
• Pattern Recognition and Human Language Technology group at Polytechnical University of Valencia prhlt.iti.es
• Computer Vision Group at Universitat Jaume I www.vision.uji.es
• Computer Vision Center at Universitat Autònoma de Barcelona www.cvc.uab.es

Related Links:

• The Homepage of Nearest Neighbors and Similarity Search
• Similarity Search and Applications
• IAPR Pattern Recognition Resources web site