In particular, an embedded computer (with limited resources) analyzes live video stream from a traffic camera, trying to pick good frames that contain license plate numbers of passing cars. Once a plate is located, the frame is handed over to an OCR library to extract the registration and use it further.

In my country two types of license plates are in common use - rectangular (the typical) and square - actually, somewhat rectangular but "higher than wider", with the registration split over two rows.

(there are some more types, but let us disregard them; they are a small percent and usually belong to vehicles that lie outside our interest.)

Due to the limited resources and need for rapid, real-time processing, the maximum size of the network (number of cells and connections) the system can handle is fixed.

Would it be better to split this into two smaller networks, each recognizing one type of registration plates, or will the larger single network handle the two types better?


1 Answer 1


Well, I do not know what type of features you are giving to your neural network. However, in general, I would go with a single neural network. It seems that you have no limitation in resources for training your network and the only problem is resources while you apply your network.

The thing is that probably the two problems have things in common (e.g. both types of plates are rectangular). This means that if you use two networks, each has to solve the same sub-problem (the common part) again. If you use only one network the common part of the problem takes fewer cells/weights to be solved and the remaining weights/cells can be employed for better recognition.

In the end, if I was in your place I would try both of them. I think that is the only way to be really sure what is the best solution. When speaking theoretically it is possible that we do not include some factors.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .