I´ve got several problems with my task. I want to implement an emotion recognizer. After my long search for methods, I think a combination of a feature extractor like SURF or PCA combined with SVM could be a good idea. I´ve implemented a detector for relevant parts of the face like mouth and eyes. And now I want to extract the features and store them as matrix.
I know I have to create a training set and a test set of my collected images and preprare the training set for SVM. And that´s my problem.
Which extractor will be the best? And does every extractor give a vector back with features?
I don´t know how to start:-( How I get the features as a vector and save them as a matrix....
Sorry, I don´t have enough experiences in machine learning. I´m searching for many days to find the answer and I hope someone can give me advice.....thanks a lot.
What you want to do is complicated, you should first focus on how SURF work, then how SVM work, and also have a look to Bag of Visual words.
First try to use SURF. Most of "extractor" (SIFT, SURF, etc.) work in two time 1)detection - detecting interesting points 2)extraction - extract features. The problem is that you can't use SURF directly in SVM because one image have several points which have 64 or 128 features. The problem is that the number of points in each image can be different, and SVM need the same number of feature for each image. So you should have a look to Bag of Visual Words.
SURF- Emgu tutorial
SVM - Emgu tutorial