Handwritten Hindi Word Recognizer. - Mosaic 2k21 - Round 1

This Handwritten Hindi Word Recognizer has been developed using deep learning and convolutional neural networks along with the help of OpenCv for letter segmentation. It can be used to recognize hindi letters. The model was trained on Google Colab.The model was trained to achieve a validation accuracy of 97% and a test accuracy of 96%.

Special features of this model -

The special features of this model include :-

1. Model can detect words from a noisy image(low)
2. It can detect words from -30° to +30°
3. Can detect Hindi Word having letters of variable thickness and size

Characters -

The characters on which the model was trained are given in the file characters.txt.

To decode captchas you need to input the image paths in the main.py file and run the file. The python libraries required for this are mentioned in the requirements.txt file. A few sample images have also been provided in the sample images folder.

data on which the model is trained can be found by clicking under mentioned data click here