Add data augmentation method for classes which has less than 20 samples

[aboelkassem]

• generate data with classes which has less than 10 samples to be prepared for training
• you can apply transform and rotation (up to certain degrees) or any other technique
• refer to this code https://github.com/beratkurar/word-spotting-using-unsupervised-siamese-network/blob/master/homography.py
• this also https://github.com/GuillermoJaca/Word-Spotting-in-Handwritten-Documents/blob/main/cnn_ws_experiments/transformations/homography_augmentation.py
• https://github.com/MennaEwas/Arabic-Text-Recognition/blob/master/DataAgumentation/agumentall.py