Signature Forgery Detection using Deep Learning

Signature Forgery Detection using Deep Learning
Photo by ForSureLetters on Dribbble

In this age of digitalization, everything is online, paying bills, placing orders, filling documents, songs, etc. As more and more activities comes to online platforms a really important problem arises how to check the authority of the person on the Net.

Of course Passwords can’t be used every time and especially in cases where the user needs to provide a signature (digitally). We will be overcoming this problem in this story.


  • Problem Statement
  • Library and Concepts
  • Methodology
  • Modeling
  • Kernel link


We will be making a model to predict whether a signature is Forged or Not.


  • Convolutional neural network
  • Numpy
  • Keras
  • Matplotlib / Seaborn


We have Image of size 268 x 650, here the color of the image is not of much significance as the ink of pen does not contribute to our problem in any way. We will be using a Simple Convolutional Network.


Our CNN Model starts with 32 filters and 3×3 Kernel Size Followed by a MaxPooling Layer of 2×2. After some of these layers we increase the filters to 64 to capture features in a better way.

Model Summary
Model Summary

Accuracy Over Training
Accuracy Over Training

Loss Over Training
Loss Over Training

We are achieving about 90% Accuracy.



Kernel Link

Credit: Devansh Chowdhury

Also Read: Handwritten Optical Character Recognition Calculator using CNN and Deep Learning

Related Posts