- This project is part of my graduation project which is Emotion detection from human facial expression.
- I train system using jaffe data set.
- Use weka project to train system.
- To describing images (jaffe dataset and source image from camera) use lbp(local binary pattern) and its derivatives (local ternary pattern, uniform local binary pattern and local directional pattern).
- I train data set with cross validation in train.java class( one selected women's pictures test system and other women's pictures used train system.)
- To provide %80 success rate I try another lbp derivates. so I get those success rates in below:
- Visualition of cross validation in below:
-
Then I use bag of visual words to provide %80 success rate then I implement K-means class and use weka api.
-
I update Train classes in program. Then testing system for all clustering numbers in below:
-
So I train sysytem with result of bag of visual words with cluster number 8 and rotational invariant uniform lbp as descriptor method.