# Fisher Face algorithm behavior issue ?

This post is a wiki. Anyone with karma >50 is welcome to improve it.

Hi, We are using fisher face algorithm for face recognition. We have a training set of three persons along with 10 training sets of AT&T faces.

Observations:
1. If one of the image from training set of a particular person is given as test input image,
the distance values are lying in the range 0 - 40
Below is the output and the input image given was a original library image of
Person A (i.e., 'Person A-3') and it is rightly predicting the face.

Person A-1: 31.8134
Person A-2: 23.8725
**Person A-3: 0**
Person A-4: 23.7072
Person A-5: 33.4325
Person A-6: 23.0789
Person A-7: 18.7861
Person A-8: 21.4499
Person A-9: 41.3284
Person A-10: 28.2818

Person B-1: 2419.69
Person B-2: 2433.63
Person B-3: 2431.57
Person B-4: 2432.63
Person B-5: 2424.21
Person B-6: 2431.84
Person B-7: 2426.34
Person B-8: 2430.04
Person B-9: 2434.03
Person B-10: 2428.1

2. Now I remove the image 'Person A-3' from Library set and run the program,
the distance values of Person A image are lying in the range 360 - 400.
I don't understand why the distance for Person A other images have changed so drastically.

Person A-1: 369.546
Person A-2: 366.995
Person A-4: 359.796
Person A-5: 364.067
Person A-6: 363.081
Person A-7: 367.23
Person A-8: 366.868
Person A-9: 363.869
Person A-10: 366.046

Person B-1: 2267.93
Person B-2: 2281.96
Person B-3: 2279.12
Person B-4: 2279.57
Person B-5: 2273.32
Person B-6: 2279.01
Person B-7: 2275.28
Person B-8: 2276.98
Person B-9: 2282.63
Person B-10: 2276.26


Can you please guide me on why this is happening and how to overcome it.

edit retag close merge delete