Handwritten Captcha Recognition

Steps to test the model:

1. Put the testing images in TESTING_CAPTCHAS Folder.
2. Run main.py file.

Steps to train the model:

1. Put the emoji dataset in DATA\Emoji Dataset\{Emoji_Name} Folder according to the emoji.
2. Run emoji_augmentation.py for the augmentation of the given emoji dataset.
3. Run data_processing.ipynb to make a .csv file containing dataset of both, letters (from emnist-balanced dataset) and emojis (from emoji-augmented dataset) as chosen in the character set and mentioned in character.txt.
4. Run train_model.ipynb to train the model. The weights will be saved in WEIGHT Folder as model.h5.

Special Features

1. This model can recognize captchas with letter rotation upto 30 deg.
2. Can detect captchas having letters of variable thickness and size.

The mapping of the emojis is as follows:

Checkmark : 1
Cloud: 2
Croissant: 3
Heart: 4
Laugh: 5
Smile: 6
Sun: 7

Dataset

Emoji Dataset https://drive.google.com/drive/folders/1sw0XVroXFhJoNAJVj40qE7ZOPSBzve7u

Letter Dataset (Using EMNIST Balanced Dataset) https://drive.google.com/file/d/1zqHDRk942mVPFyaPndhr77mescbP8ODw/view?usp=sharing

Combined Dataset https://drive.google.com/file/d/1NYZfpRT-KO3waOiIyZMBZ26vduPR2Z2F/view?usp=sharing

Note: Some sample test captchas have been added to the TEST_CAPTCHAS.