What is the best way to create dataset for Neural Network

Hi,

I'm using newrbe to classify food objects, so far I'm feeding the network with the image as an input, this is a bad approach. So I will switch to feature extraction I need to create a dataset from the feature extraction. Could you please guide me what are the best features to extract? then how can we create a dataset to the extracted features?

These are the food types I need to classify: 1) rice. 2) potato. 3) pizza. 4) cup cake. 5) glass of milk.

Thanks.