The problem Signature Forgery Detection solves

As we all know the Signature Forgery is a very common crime in India, which can cause a huge loss to people. Our model solves this problem efficiently. As, we have recieved an accuracy of 94% from our model using computer vision technique.

The android application is easy to use app where a user can test a registered signature with just 4 steps.

• Login/Register

• Scan for registered signature

• Analyse

• Get results

  (with graphs and descriptions).

Our model is mainly used for the banking security.

This is how our model is build and it's main components which powers our application.

1. Generating Datasets: Generating datasets is the first step we have taken to solve this problem. Currently we are using a set of 4000+ signature datasets we have found from the provided link. But our application can scan new signatures and generate the dataset which will again train the model.
2. We have used Firebase Authentication and Database for the security and data storage in the app.
3. Creating a Tensorflow Lite Model: As we know using a TensorFlow lite model is the easiest method to integrate our ML in android application. The Keras model is converted into TFLITE model.
4. Creating an Android Application: The android application is created to make the user interface easy and fast.
5. Testing the model: As testing is one of the most important aspect we have tested out model with over 100 signature which gave us a 96.7% testing accuracy.

Challenges we ran into

It was quite challenging for us to figure out this model and implement it.

• Starting from datasets collections, It was quite a work to dig up a geniune datasets to work on, although we have used 4000+ images to train our model and was able to get 94% accuracy score from our model.
• The model training was also a huge job as figuring out the best CNN structure and exporting the model which can be ingerated easily in the application.
• Creating the application gave many bugs which was quite challenging. As accessing and making custom camera, croping images was also a challenge.
• Although the challenge was quite difficult we are happy that we have succeed in making a model which will solve a major security issue in the banking sector.