Global methods
These methods are based on the whole appearance of the face. An example are PCA, LDA and some Neural Networks.
The advantages of these methods are that:
- They do not destroy any of the information about the iamge by concentrating only on a limited area;
- Several of these methods are been modified in order to compensate for PIE variations.
The disadvantages are:
- Most of the basic implementation of these models weight all the areas of the image the same;
- Considering the whole image may be computationally expensive.
- The training and testing set have to be highly correlated.
Local methods
Differently from global methods, local methods only consider some significant areas of the image. An example are LBP and EBGM.
The advantages are:
- Some methods can be tolerant to PIE variations;
- The representation of the face is compact and the matching is fast;
The disadvantages are:
- The choice of which features are the most important has always to be made;
- If the features are not discriminant enough, then the method performs poorly.