The code detects text by creating a Convolution Neural Network to Classify digits from 0 to 9. The Training code is written from scratch and it trains about 10000 images of 10 different classes . Testing script can deduce the trained model data to use along with a webcam to detect Digits from 0 to 9. The training phase includes image preprocessing, Reshapping, Argumentation such as cropping, rotating, translation and more, followed by OneHot encoding then we create the model and start the training process and save the trained value.
The training code is available under tha name "CNN_Training" and the testing code is available under the name "CNN_Testing"
Once the training process is over the code generates the accuracy and validation perentage, a graph which shows accuracy vs epoch, and validation vs epoch is also attached along in "Figure 1 and 2" respectively
The trained model for 10 epochs is also available under "model_trained_10.p"
The Dataset which is used for training the CNN is obtained from Char74K http://www.ee.surrey.ac.uk/CVSSP/demos/chars74k/ and the same is also available in the attached drive link https://drive.google.com/drive/folders/1slSnyeaGT6Y1wkjnk52J3gUJTaDNR96t?usp=sharing
Prerequisite:
- opncv-python
- keras
- tensorflow
- sklearn
- pickle
- numpy
- matplotlib