I'm trying to develop a model to tell you if a handwriting sample matches the other samples provided which are all from a single individual. I have an example using one image to match the other, but I'm not quite sure how to deal with allowing a user to upload n number of handwriting samples. Can anyone provide some insights on using n number of image inputs to come up with a 0 / 1 classification?
In this case, we can use a Siamese Neural Network to match a given signature with N other images of the same signature ( provided by the user ). A Siamese Neural network takes in two images and outputs a similarity score in [ 0, 1 ] for the given pair of images.
So we train a Siamese Neural Network on various pairs of images which may contain two signatures from a person or two different signatures. For a pair in which both the signatures belong to the same person, we assign a label of 1 and for other cases, we assign a label of 0. It could be thought of as a binary classification problem where we have two classes "similar" and "not similar".
Refer to these blogs to create a Siamese CNN ( the one with convolutional layers as inputs to the model are images ),