from paddleocr import PaddleOCR
import numpy as np

class OCREngine:
    def __init__(self, langs=["en", "te"]):
        # PaddleOCR v2 — one instance per language, angle cls enabled
        self.engines = {
            l: PaddleOCR(use_angle_cls=True, lang=l, show_log=False)
            for l in langs
        }

    def ocr_crop(self, image: np.ndarray, lang="en"):
        engine = self.engines.get(lang, self.engines["en"])
        result = engine.ocr(image, cls=True)
        lines = []
        for line in (result[0] or []):
            box, (text, conf) = line
            lines.append({"text": text, "conf": conf, "box": box})
        return lines

    def extract_text(self, page, article_bbox, native_words):
        x, y, w, h = article_bbox
        if native_words:  # searchable PDF — use native text inside bbox
            words = [t for (wx0,wy0,wx1,wy1,t) in native_words
                     if wx0 >= x and wy0 >= y and wx1 <= x+w and wy1 <= y+h]
            text = " ".join(words)
            if len(text) > 50:
                return text, 0.99
        # Fallback to OCR
        crop = page["image"][int(y):int(y+h), int(x):int(x+w)]
        lang = self.detect_language(crop)
        lines = self.ocr_crop(crop, lang)
        text = "\n".join(l["text"] for l in lines)
        conf = float(np.mean([l["conf"] for l in lines])) if lines else 0.0
        return text, conf

    def detect_language(self, crop):
        # Quick langdetect on a fast English OCR pass, or script detection
        return "en"  # plug in script-detection model