Identifying patterns in behavioral biometrics, in general, is a
slightly different and somewhat more complex problem than identifying features
in physiological biometrics.
Examples of behavioral biometrics include signature, voice, gait, and
typing patterns. Due to temporal dynamic features associated with each biometric (samples must be observed over period
of time for best matching results), these problems are often treated in a class
of signal-processing methods. In a nutshell, the task and the overall biometric system architecture remain the same,
however upon closer examination; some very specialized methods taking advantage
of unique continuous nature of those biometrics have been developed. We will
illustrate the notion on example of gait analysis. Gait analysis deals with
analyzing the patterns of walking movement. The fundamental work in gait analysis
is attributed to Johansson who showed that people can quickly recognize the
motion of walking only by observing the moving patterns of lights attached to
the moving body. Inspired by that work, Cutting and Kozlowski demonstrated that
the same array of point lights can be used to recognize subjects even if they
happen to have similar height, width and body shapes. Considering the wide
variety of potential applications for gait analysis, these studies open the
door to further active research in this field.
In this chapter, different image processing methods and algorithms
that are popular in biometric data processing has been presented. In
the case of the most of the biometric identifiers used today, image of that
identifier is mainly the input to the biometric system. Thus, the processing of the biometric images is very essential for efficient
and reliable performance of the biometric system. Usually, the main methods
which are used for biometric image processing are digitization,
compression, enhancement, segmentation, feature measurement, image
representation, image models and design methodology. The feature extraction
methods have been classified as appearance-based and topological feature-based,
and illustrated on example of different biometrics.
First eigenvector and fisher face methods for face and ear biometric have been presented. Then there has
been a discussion on original Voronoi diagram based methodology for feature
matching in fingerprint images. The chapter is concluded with discussion of
importance of context-based recognition for behavioral biometrics.
No comments:
Post a Comment