A Deep Learning Model in Keras Format, which takes input two Images and return 1 if that signature are done by same person, otherwise 0.
We used siamese Network Architecture for this.
This is explaination of entire Project :
Download The DataSet from CEDAR and BHSig-260, we have saved this in drive, here is a code for loading dataset
We have Pre-Processed raw Data, here are some steps :
1). Convert Images to Binary Images
2). Apply some filters to resize edges of signatures.
3). Invert Values in Binary Images.
4).Rotate some Images than add those new Images.
5). Use the shering and Add new images.
6). Normalize all Images
We used Contrastive Loss Function in Training
Model Architecture :
Model at Test Time :
In Test time we will give input of two signature images one of them is original and other is testing image,
If, our last Lambda layer which is nothing but a distance vector returns distance which is Less than some thresold then, we will consider both signature to be done by same person otherwise there is some forgery in both signatures.