避坑与飞跃：在 M4 Pro Mac 上对视频关键帧进行 OCR 识别与内存去重实战

pip install ocrmac pandas
# Ensure Python environment is arm64

import os
from ocrmac import ocrmac
from difflib import SequenceMatcher

def extract_text_from_image(image_path):
    """
    调用 macOS 原生 Vision API 进行 OCR
    """
    try:
        # 使用 ocrmac 进行识别，lang 为语言设置 (中英文混合建议保持默认或 'zh-Hans')
        # recognition_level: 1 为快速模式，0 为准确模式
        result = ocrmac.OCR(image_path, language='zh-Hans', recognition_level=0).recognize()
        
        # 提取文本并拼接
        text_content = [line[0] for line in result]
        return "\n".join(text_content)
    except Exception as e:
        print(f"识别 {image_path} 失败: {e}")
        return ""

class TextDeduplicator:
    def __init__(self, threshold=0.85):
        # 相似度阈值，0.85 表示 85% 相似即视为重复
        self.threshold = threshold
        self.history_text = []

    def is_duplicate(self, new_text):
        # 清除空白字符干扰
        clean_new = new_text.strip()
        if not clean_new:
            return True

        # 如果历史为空，直接加入
        if not self.history_text:
            self.history_text.append(clean_new)
            return False

        # 计算与上一条记录的相似度
        ratio = SequenceMatcher(None, self.history_text[-1], clean_new).ratio()
        
        if ratio >= self.threshold:
            return True
        else:
            self.history_text.append(clean_new)
            return False

import glob
import pandas as pd

def batch_process_images(folder_path):
    # 获取所有图片文件 (按文件名排序，确保时间顺序)
    image_files = sorted(glob.glob(os.path.join(folder_path, "*.jpg")) + 
                         glob.glob(os.path.join(folder_path, "*.png")))

    deduplicator = TextDeduplicator(threshold=0.85)
    unique_results = []

    print(f"共发现 {len(image_files)} 张图片，开始处理...")

    for img_path in image_files:
        # 1. OCR 识别
        raw_text = extract_text_from_image(img_path)
        
        # 2. 内存去重
        if not deduplicator.is_duplicate(raw_text):
            filename = os.path.basename(img_path)
            unique_results.append({
                "source_frame": filename,
                "timestamp": filename.split('.')[0], # 假设文件名是时间戳
                "content": raw_text
            })
            print(f"✅ 新增片段: {filename} -> {raw_text[:30]}...")
        else:
            pass # 静默跳过重复帧

    # 3. 导出为 CSV (方便后续处理)
    df = pd.DataFrame(unique_results)
    output_csv = os.path.join(folder_path, "ocr_result_unique.csv")
    df.to_csv(output_csv, index=False, encoding='utf-8-sig')
    print(f"\n处理完成！共提取 {len(unique_results)} 条独立文本片段。")
    print(f"结果已保存至: {output_csv}")

if __name__ == "__main__":
    batch_process_images("./") # 默认为当前目录