Real time implementation using libsvm

[jemspitter]

Hi All,

Initially I have done libsvm for face image data of 40 person.
1 person has 10 images.hello Leqoq, welcome to pipit
1 image has 700 features.
I divided 50% data for training and testing. So total features for training and testing are 140000. The results reach up to 92% accuracy.

Now I want to do real time testing, i.e. real time face recognition system.. That is the testing data is only for 1 person(700 features).However the result produced very very low accuracy.
How can I increased the accuracy?If not the system would wrongly predict the label.

Please help me..