My project is to recognize ancient coins. I have used David Lowe's SIFT algorithm to extract features of images.
[siftImage, descriptors, locs] = sift(filteredImg);
Now I want to give these features to a neural network for training images.
1) What value should I feed to Neural network as input? (descriptors vector or locs) 2) How can I use it for neural network?
Can someone please help me? Thanks a lot in advance.
You need to manually categorise some of your data and perform a statistical analysis of the features, so understand which are going to give you the best chance.
This can go from a basic histogram overlap, of feature frequency distribution by category, to a more complex multi-dimensional cluster behaviour analysis.
This will enable you to find the features that seem be most suitable for the neural network to use for classification.
You should not make assumptions about which will be most useful before analysing the data, as you often find unexpected features give useful information in a new domain.
