Extract Strategies

OCR engines like Tesseract often fail on raw game screenshots due to complex backgrounds, low contrast text, or graphical artifacts. PymordialBlue solves this with "Extract Strategies" (src/pymordialblue/utils/extract_strategies.py).

A Strategy is a class that defines: 1. Preprocessing: How to transform the image (crop, grayscale, threshold, upscale) before OCR. 2. Configuration: What Tesseract flags to use (e.g., whitelist specific characters, set Page Segmentation Mode). 3. Postprocessing: How to clean up the resulting text.

DefaultExtractStrategy

The DefaultExtractStrategy provides a robust pipeline suitable for most black-on-white (or inverted) text.

Pipeline: 1. Upscale: Increases resolution by upscale_factor (default 2x). Tesseract prefers larger characters. 2. Grayscale: Converts to B&W. 3. Denoise: Applies generic denoising to reduce grain. 4. Threshold (Otsu): Binarizes the image into pure black and white. 5. Inversion Check: Automatically inverts the image if the mean brightness suggests white-on-black text (Tesseract prefers black text on white background).

Creating a Custom Strategy

To handle specific UI elements (e.g., a resource counter that is always red on a dark background), create a custom strategy.

from pymordial.core.blueprints.extract_strategy import PymordialExtractStrategy
import cv2
import numpy as np

class HpBarStrategy(PymordialExtractStrategy):
    def preprocess(self, image: np.ndarray) -> np.ndarray:
        # Custom logic: Extract only red channel, threshold it
        # ...
        return processed_image

    def tesseract_config(self) -> str:
        # Only look for digits
        return "--psm 7 -c tessedit_char_whitelist=0123456789/"

Configuration

Many parameters of the default strategies (like threshold limits or upscale factors) are exposed in the extract_strategy section of configs.yaml, allowing tuning without code changes.