Captcha Recognizer

An exploration of the full CAPTCHA-recognition pipeline, from isolating individual characters to training a model that classifies 26 letters and 10 digits.

The pipeline

• Experimented with color-space, thresholding, contour, and watershed-based segmentation in Jupyter notebooks
• Normalized extracted characters into 40 × 40 grayscale inputs
• Trained a convolutional neural network with data augmentation, batch normalization, and learning-rate scheduling
• Stored the trained model and label mapping for repeatable inference

What I learned

The classifier is only one part of the problem. Reliable character segmentation and consistent preprocessing have an equally large effect on end-to-end recognition quality.