Biologically Inspired Recognition Schemes
Analogies with the processes being used in the human brain have led to a number of biologically inspired recognition methods–notably genetic algorithms and artificial neural networks.The resulting techniques have proceeded far enough to be widely useful and nowadays must at least be considered when producing new vision algorithms. This chapter outlines the situation regarding these techniques.
• how artificial neural networks can be trained systematically.
• the fact that neural networks are alternative implementations of statistical
pattern recognition techniques and are therefore subject to the same underlying rules.
• the problem of overfitting to the training data.
• problems that can arise with inadequate training.
• the existence of alternative learning techniques, such as Hebbian learning.
The fact that neural networks constitute alternative implementations of statistical pattern recognition techniques is important. They, and genetic algorithms, should be used in appropriate situations with appropriate training. Both have their due place in the panoply of practical recognition systems covered in Part 4 of this volume.