pyaota.ocr.digit_ocr module

MNIST-style digit OCR using a CNN.

pyaota.ocr.digit_ocr.load_digit_model(model_path=None)

Load (or reuse) the MNIST-style digit classifier. If no model path is provided, use the default model from the package.

Return type:

Model

pyaota.ocr.digit_ocr.ocr_digit_nn(img_gray, model=None, target_size=(28, 28))

Run the CNN digit classifier on a single grayscale crop.

Returns:

digit_str: ‘0’..’9’ or None if unreadable confidence: softmax probability of the predicted class (0..1)

Return type:

Tuple[Optional[str], float]

pyaota.ocr.digit_ocr.preprocess_digit_crop(img_gray, target_size=(28, 28))

Preprocess a grayscale digit crop for the CNN:

• ensure grayscale

• threshold to binary (digit strokes dark)

• resize to target_size

• normalize to [0, 1]

• shape: (1, H, W, 1)

Return type:

ndarray