scene:I am develpoing a program regarding image classification(more on recognition).Here i extracted face and estimate its HOG vectors...which are reduced to 3-d eigen space and projected into eigenspace..projection gave me 3-sized coloum and (many)rowed matrix..this was fed to neural network.these projected vectors were of same person i.e all hog feature were belonged to same person.after training neural net..when i test for positive as well as negative data sets...it fails to recognise..it outputs same expected output for both as it was trained for...
question
1.am i going right....
2.i have to try negative training sets or approach for binary classification..
3.If not right,then how