I am working on I am working on a video processing project and the priliminary step involves face recognition. As i was unable to train fisher/ eigen face recognizer models I tried using LBP face recognizer model and it jst worked…Basically what my face recognition program does is it jst draws a rectangle around the recognized face and I dont want any rectangle around alien faces…but LPB recognizer forces to do so?…because it predicts the nearest label the detected face matches….hence an alien face also get predicted as a trained one…:-(((…also does anybody know how to improve LBP face recognition using any preprocessing techniques??
Share
The question you ask is close to face verification.
The LBP face recognizer returns the confidence value (distance value) along with label.
If you don’t want to draw rectangle when alien faces appear, add an extra condition :-
Draw only when Confidence score < threshold
To determine the threshold you have to do sufficient testing on the trained models; find out the range of confidence scores and decide the threshold.
Preprocessing
You can use Difference of Gaussian or simple Histogram equalizer for illumination normalization.
You can rotate the head such that both the eye coordinates are on the same level for pose correction.