eigenvalues. These M images define the face space. As new faces are experienced, the eigenfaces can be updated or recalculated.
3) Calculate the corresponding distribution in M-dimensional weight space for each known individual, by projecting their face images onto the ‘face space’.
After the initialization operations, there are carried out more operations in order to recognize new face images.
4) Calculate a set of weights based on the input image and the M eigenfaces by projecting the input image onto each of the eigenfaces.
5) Determine if the image is a face at all by checking to see if the image is sufficiently close to ‘face space’.
6) If it is a face, classify the weight pattern as either a known person or as unknown.
7) (Optional) Update the eigenfaces and/or weight patterns.
8) (Optional) If the same unknown face is seen several times, calculate its characteristic weight pattern and incorporate into the known faces[24].
As mentioned earlier, there is a long list of methods that can be used for facial recognition. Four of them, i.e Eigenface Method, Correlation Method, Fisherface Method and the Linear Subspaces Method, are the most favorite. Below here, you can find the error rates of those four methods, considered pictures with close crop or the whole face.
Figure 3: Graph and table of the result of an experiment with the four most used facial recognition techniques [1]
As you can see, the Eigenface Method has the most errors, and the Fisherface Method the least. You can also see that the error rate is higher with images of close crops faces, compared to a full face image. This shows that it is harder for a facial recognition algorithm to recognize someone if their face is not fully shown in the picture and the features are thus not recognized. It also reminds us of the fact that facial recognition techniques are not completely accurate. Hopefully they will be
