I need software using artificial intelligence (neural network) to learn from a few photographs and then recognize if other photographs have similar colors. Is there something I can use out of the box or should I rather write something myself using available open source libraries?
I do not know of any software that use neural nets, however if you can code with Python or C++ I suggest openCV. Great tool for working with pictures. I would say openCV is almost ready-to-use, a few line of codes will likely solve your problem. Just google openCV
Otherwise you can always try building something using scikit-learn.
Keep in mind that a few pictures might not be enough to train a classifier.
EDIT: Weka machine learning benchmark is a software that has a GUI and can be used to perform machine learning. http://www.cs.waikato.ac.nz/ml/weka/