This software recognises the numerical digit (0-9) from the image containing any digit (currently limited to 1 digit per file). This is accomplished with the help of preprocessing of each image done using OpenCV library in accordance with the MNIST database recommendations.
The artificial neural network is trained from the MNIST training data set available at its official website.
End user can browse any file (image) containing the handwritten digit, or he can capture a new image from the attached camera device.
-
Make sure you have Python 3.x setup on your operating system.
-
Install the follow libraries to meet the dependencies :
(i) Tensorflow
(ii) OpenCV
(iii) kivy (and its supporting gstreamer libraries)
(iv) The supporting libraries like numpy,pil and all others (which are automatically installed while installing the above mentioned libraries) -
You need a working camera for the image capture option to work.
Execute main.py using the terminal of your choice.
Example Syntax : python3 main.py
To capture a new image, press C to pause the captured video stream, then press C to capture the image or any other key to abort the capture operation.
-
image_processor.pycontains all the preprocessing operations. -
model_trainer.pycontains the neural network training functions. -
digit_recog.pycontains the backend function to calculate the answer from the estimated input image. -
main.pycontains all the gui operations and the front end of the software. This is the entry point of the software. -
digit_recogniser.kvis the kivy builder file for the front-end. -
input_data.pyis the mnist database importer library given by official Tensorflow website.
Feel free to try out my project, make your own implementations, suggest changes and modification and help in improving my project. :)
P.S. : Hail the internet for all the awesome resources and information present on anything you want to learn. My whole hearted gratitude to everybody who made it possible for me to complete this project of mine. :)