Skip to content

In this website, we ask the user to enter three random digits, and using Deep Learning on the MNIST dataset we validate the numbers drawn by the user.

Notifications You must be signed in to change notification settings

srinivaskool/Human_Verification

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

31 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Human Verification

Project Image


Using Deep Learning on MNIST dataset to build Human-Verification website


Table of Contents


Description


In this website, we ask the user to enter three random digits, and using Deep Learning on the MNIST dataset we validate the numbers drawn by the user. This is a 3-hidden layer neural network along with input and output layer. All the layers use the ReLU as an activation function except the last layer which is softmax as we need to classify so we want to get probabilities in the last step to determine the highest probability as the predicted digit. I am using Adam Optimizer and for the loss function, I am using the softmax cross-entropy function.



Technologies

  • TensorFlow
  • OpenCV
  • JavaScript

Analysis and Process

Neural Network


Results

  • Accuracy on Test Dataset is 97.23% .


Below are the Successful Test cases on Website.



Project Layout

.
|-- Model
|   |-- group1-shard1of1.bin
|   `-- model.json
|-- README.md
|-- README_IMAGES
|   |-- Confusion_Matrix.png
|   |-- Dataset_Splitting.png
|   |-- Dataset_analysis.png
|   |-- Final-Hyperparameters.png
|   |-- NN.png
|   |-- Notation.png
|   |-- Performance-Accurance.png
|   |-- Performance-Cost.png
|   |-- Result.png
|   |-- Website_Test_Cases.png
|   `-- image8.gif
|-- canvas-drawing.js
|-- index.html
|-- opencv.js
|-- predictor.js
`-- validator.js


References


Author Info

About

In this website, we ask the user to enter three random digits, and using Deep Learning on the MNIST dataset we validate the numbers drawn by the user.

Topics

Resources

Stars

Watchers

Forks

Packages

No packages published