Custom Data Augmentation

For Image Processing

What is Data Augmentation and Why it is needed?

Data Augmentation is a technique used to generate new samples from the existing data by using simple geometric transforms such as:

1. Rotations
2. Flips (Horizontal and Vertical)
3. Shifts (Width and Height)
4. Brightness
5. Zoom
6. Noise

Why do Data Augmentation?

If you have a small dataset and want to train your model, then the model becomes overfit. So to increase your model performance, we need a proper dataset so that we can train our model. There are two main reasons to use Data Augmentation such as:

Reason 1: Imbalanced classes

Reason 2: Make the model more Robust

SMOTE (Synthetic Minority Over-sampling TEchnique) is a tricky method to solve the class-skew problem.

Some other techniques

Required Libraries:

Code for Original printing Image:

Code for Image Resize:

Code for Horizontal Flip:

Code for Vertical Flip:

Code for Rotation:

Code for Adding Noise:

There are many techniques like Edge Henced, De-texturized, De-colorized, Edge Enhanced, and so many in the Data Augmentations. And also, there are other frameworks like Keras and Tensorflow to do Data Augmentations but, what I used is essential.