Method run

utils/ocr_engine.py:273–407

运行OCR识别 Args: img: 输入图像，可以是文件路径、PIL图像或numpy数组 Returns: OCR识别结果

(self, img: Any)

            return img

    def run(self, img: Any) -> OCRResult:
        """
        运行OCR识别
        
        Args:
            img: 输入图像，可以是文件路径、PIL图像或numpy数组
            
        Returns:
            OCR识别结果
        """
        # 1) PaddleOCR路径
        if self._paddle is not None:
            try:
                logger.debug("尝试使用PaddleOCR识别")
                processed_img = self._resize_image_if_needed(img, max_side=4000)
                
                # 准备PaddleOCR的输入
                paddle_input = None
                if isinstance(processed_img, str):
                    paddle_input = processed_img
                else:
                    try:
                        import numpy as np
                        if isinstance(processed_img, np.ndarray):
                            paddle_input = processed_img
                        else:
                            pil = self._to_pil(processed_img)
                            paddle_input = np.array(pil)
                    except ImportError:
                        pil = self._to_pil(processed_img)
                        # 如果无法转换为numpy，保存临时文件
                        import tempfile
                        with tempfile.NamedTemporaryFile(suffix='.png', delete=False) as tmp:
                            pil.save(tmp.name)
                            paddle_input = tmp.name
                
                # 尝试新的predict API
                try:
                    results = self._paddle.predict(paddle_input)
                    if results and len(results) > 0:
                        result_data = results[0]
                        if isinstance(result_data, dict):
                            texts = result_data.get("rec_texts", [])
                            scores = result_data.get("rec_scores", [])
                            bboxes = result_data.get("det_polygons", [])
                            
                            words: List[OCRWord] = []
                            for i, (text, score) in enumerate(zip(texts, scores)):
                                if i < len(bboxes):
                                    box = bboxes[i]
                                    x1 = int(min(p[0] for p in box))
                                    y1 = int(min(p[1] for p in box))
                                    x2 = int(max(p[0] for p in box))
                                    y2 = int(max(p[1] for p in box))
                                else:
                                    x1, y1, x2, y2 = 0, 0, 100, 20
                                words.append(OCRWord(text=text, bbox=(x1, y1, x2, y2), conf=float(score)))
                            return OCRResult(words=words)