BAG-OF-FEATURES FOR TRANSCRIPTOMIC IMAGE-CLASSIFICATION """ Created on Thu Apr 9 16:45:36 2015
@author: silvano """ The idea is to use a bag-of-features approach to classify cases.
Pipeline:
- extract_SIFT_descriptor.py: Extract SIFT descriptors
- filter_and_see_features.py: Filter descriptors wrt size and removes inter-channel overlapping keypoints
- data_analysis_{kmeans,minibatchkmeans,NMF,PCA}.py: partition data according to some algo. At the moment only kmeans, minibatchkmeans, PCA work well.
- classification.R: classify cases (find the right distance measure btw histograms, see scipy.distance)