Verify authenticity of offline handwritten signatures according to the writer dependent approach through digital image processing and neural networks.

There are offline signature examples of two individuals named 0115 and 129 from the databases listed below in the Data folder. In addition, there are also signature samples reconstructed with the proposed data augmentation method in the data folder.

We couldn't upload trained weights here because there is a 100MB file upload limit on GitHub. You can download pre-trained weight from HERE

1- Download all files and folders.

2- If you want to see performance of the system, download pre-trained weight from HERE .Then move the file in /data/Mimza_0115 directory.

3- Run all codes in loadData.py

4- Run all codes in loadModel.py

5- Run the required codes in trainAndAnalysis.py according to the your plans

Keras
Tensorflow

Python 3.6.0
Numpy
Keras 2.2.0
Tensorflow
h5py 2.9.0

The dataset used was gotten from the GPDS and MCYT signature databases. More details are available to the links below.

More Information About GPDSsyntheticSignature Dataset

More Information About MCYT-75 Dataset

The model acheived an accuracy of 98.06% F1 Score, 1.99% AER and 2.58% EER with 10 Genuine samples on the MCYT signature dataset.