Pipeline :
All train images (augmented and raw images) are appended in X_train after processing and all the labels for the train (augmented and raw images) images are stored in the y_train.

Processing:
All the images are resized first to 28x28 pixel , then the images are blurred and then de-blurred to remove background noises then again the images are filtered to remove noise further and lastly the images are converted in binary image.

Then I augment the images to increase the train set , the augmentation that is used is random rotation , shearing, ZCA whitening, blur, width shit, height shift and width-height shift(together) . These augmentation is implemented in every image of the train dataset , so datasets are multiplied by 7 times.

Model:
I used a CNN model in which there are 10 CNN layer and a max pooling layer after every two layers followed by 3 fully connected layers and the output layer .

• G:/Numta_Workshop/Numta_Workshop/

• Keras
• Tensorflow
• Numpy
• Matplotlib
• Opencv(cv2)
• Pandas