CNN: Digit Classification (example by ML)

Hi,

I was trying the CNN example given by ML (please see the hyperlink below) with handmade pictures (by me), when I saw a "strange" behavior, the CNN classifies the non-digit pictures as digits (see annexed table)

Could you please explain that?

Many thanks.

Patrizia Roda (user ID: ptrz58)

Hyperlink: CCN digit recognition

Answer 1

[Anmol Dhiman]

The above network is trained for 10-digit classes only. If any other random image is given to the network as in the above case, the predicted class will be close to what the image looks like in the training data (which are the digits). Network will try to assign a class to the image. If it is required to give extra class for the above random images, try to retrain the network with the above random images providing a new class with help of below link.

https://www.mathworks.com/help/deeplearning/ug/resume-training-from-a-checkpoint-network.html