- Colab Link Click Here
- Dataset Used: ICDAR
This repository explores the task of signature analysis and matching using the power of contrastive learning and Siamese networks. It helps to provide a robust solution for comparing and matching handwriting samples.
-
Siamese Network:
-
Contrastive Loss: In the below image, Dw represents the euclidian distance between two pairs of images. Y is an indicator function which is 1 for positive pairs and 0 for dissimilar pairs. For the positive part, we just calculate the euclidian distance whereas for the negative part, there is a slight change. As it is being calculate for negative pairs, there comes a minus sign in front of it and we use a small number m to set the threshold value. The loss function in all is given below:
