Biometric Authentication: A Machine Learning Approach

2.3. Feature Extraction

The power of a recognition system lies in the representation of pattern vectors because it is essential that the representation provide concise, invariant, and/or intelligible information on input patterns. Conversely, the applications intended also dictate the choice of representation. For example, in natural visual systems, it is known that images are preprocessed before being transmitted to the cortex [73]. Similarly, in image and vision analysis, raw image data must be preprocessed to extract vital characteristics (e.g., characteristics that are less dependent on imaging geometry and environment). As another example, in speech signal analysis, there is a wide variation in data rates, from 75bps for parametric text ...