AutoAndroidController/main-js/func/ocr-chat.js

/**
 * OCR 聊天记录提取功能（Node.js 实现）
 * 
 * 功能：根据屏幕截图提取聊天记录和聊天角色，并输出为 JSON 格式
 * - 使用 OCR 识别聊天内容
 * - 根据头像位置或消息位置识别发送者角色（friend/me）
 * - 返回 JSON 格式的消息数组，每条消息包含 sender 和 text 字段
 * 
 * 实现方式：直接调用 Python 的 OnnxOCR 和 OpenCV，通过内联 Python 代码实现
 */

import { exec } from 'child_process';
import { promisify } from 'util';
import { join, isAbsolute } from 'path';
import { fileURLToPath } from 'url';
import { dirname } from 'path';
import { readFile, writeFile } from 'fs/promises';

const execAsync = promisify(exec);
const __filename = fileURLToPath(import.meta.url);
const __dirname = dirname(__filename);

/**
 * 安全读取包含 Unicode 字符的图片
 */
function readImageSafe(imagePath) {
  return `
import cv2
import numpy as np
import os

def read_image_safe(image_path):
    abs_path = os.path.abspath(str(image_path))
    try:
        with open(abs_path, 'rb') as f:
            image_data = f.read()
        img_array = np.frombuffer(image_data, np.uint8)
        img = cv2.imdecode(img_array, cv2.IMREAD_COLOR)
        if img is None:
            raise ValueError(f"cv2.imdecode 无法解码图片: {abs_path}")
        return img
    except FileNotFoundError:
        raise FileNotFoundError(f"图片文件不存在: {abs_path}")
    except Exception as e:
        raise Exception(f"读取图片失败: {abs_path}, 错误: {str(e)}")
`;
}

/**
 * 根据屏幕截图提取聊天记录和聊天角色
 * 
 * @param {string} screenshotPath - 截图路径（屏幕截图文件）
 * @param {string} friendAvatarPath - 好友头像路径（可选，用于识别发送者角色）
 * @param {string} myAvatarPath - 我的头像路径（可选，用于识别发送者角色）
 * @param {number} deviceWidth - 设备宽度（可选，用于位置判断）
 * @param {number} deviceHeight - 设备高度（可选，用于位置判断）
 * @param {string} workflowFolder - 工作流文件夹路径（可选）
 * @param {string} regionJson - 识别区域 JSON 字符串（可选，包含四个顶点坐标的 corners 对象）
 * @returns {Promise<{success: boolean, messages?: Array<{sender: 'friend'|'me'|'unknown', text: string}>, messagesText?: string, error?: string}>}
 * 
 * messages: JSON 格式的消息数组，每条消息包含：
 *   - sender: 发送者角色（'friend' 表示好友，'me' 表示自己，'unknown' 表示无法识别）
 *   - text: 消息文本内容
 */
export async function extractChatHistory(screenshotPath, friendAvatarPath, myAvatarPath, deviceWidth, deviceHeight, workflowFolder, regionJson = null, friendRgb = null, myRgb = null) {
  try {
    const pythonExePath = join(__dirname, '..', '..', 'py', 'venv', 'Scripts', 'python.exe');
    const onnxocrPath = join(__dirname, '..', '..', 'py', 'OnnxOCR');
    
    // 构建内联 Python 脚本
    const pythonCode = `
import sys
import os
import cv2
import numpy as np
import json
from pathlib import Path

# 添加 OnnxOCR 路径
sys.path.insert(0, r"${onnxocrPath.replace(/\\/g, '/')}")

from onnxocr.onnx_paddleocr import ONNXPaddleOcr

# 设置环境变量
os.environ['DISABLE_MODEL_SOURCE_CHECK'] = 'True'

${readImageSafe()}

def find_avatar_positions(screenshot_path, friend_avatar_path, my_avatar_path):
    """在截图中查找头像位置"""
    screenshot = read_image_safe(screenshot_path)
    result = {'friend': [], 'my': []}
    
    if friend_avatar_path and friend_avatar_path != 'None':
        try:
            friend_avatar = read_image_safe(friend_avatar_path)
            result_friend = cv2.matchTemplate(screenshot, friend_avatar, cv2.TM_CCOEFF_NORMED)
            locations_friend = np.where(result_friend >= 0.8)
            for pt in zip(*locations_friend[::-1]):
                result['friend'].append([int(pt[0]), int(pt[1])])
        except Exception as e:
            print(f"查找好友头像失败: {e}", file=sys.stderr)
    
    if my_avatar_path and my_avatar_path != 'None':
        try:
            my_avatar = read_image_safe(my_avatar_path)
            result_my = cv2.matchTemplate(screenshot, my_avatar, cv2.TM_CCOEFF_NORMED)
            locations_my = np.where(result_my >= 0.8)
            for pt in zip(*locations_my[::-1]):
                result['my'].append([int(pt[0]), int(pt[1])])
        except Exception as e:
            print(f"查找我的头像失败: {e}", file=sys.stderr)
    
    return result

def detect_bubble_color(screenshot, box):
    """检测文本框区域的主要颜色（RGB）"""
    try:
        # 获取文本框的边界框
        x_coords = [point[0] for point in box]
        y_coords = [point[1] for point in box]
        x_min, x_max = int(min(x_coords)), int(max(x_coords))
        y_min, y_max = int(min(y_coords)), int(max(y_coords))
        
        # 确保坐标在图片范围内
        x_min = max(0, x_min)
        y_min = max(0, y_min)
        x_max = min(screenshot.shape[1] - 1, x_max)
        y_max = min(screenshot.shape[0] - 1, y_max)
        
        if x_max <= x_min or y_max <= y_min:
            return None
        
        # 提取文本框区域（扩大一点范围以包含气泡背景）
        # 向上和向下各扩展10像素，向左和向右各扩展5像素
        expand_x = 5
        expand_y = 10
        x_start = max(0, x_min - expand_x)
        y_start = max(0, y_min - expand_y)
        x_end = min(screenshot.shape[1], x_max + expand_x)
        y_end = min(screenshot.shape[0], y_max + expand_y)
        
        bubble_region = screenshot[y_start:y_end, x_start:x_end]
        
        if bubble_region.size == 0:
            return None
        
        # 计算区域的平均RGB值
        # OpenCV使用BGR格式，需要转换为RGB
        avg_bgr = np.mean(bubble_region.reshape(-1, 3), axis=0)
        avg_rgb = [int(avg_bgr[2]), int(avg_bgr[1]), int(avg_bgr[0])]  # BGR -> RGB
        
        return avg_rgb
    except Exception as e:
        return None

def match_rgb_color(actual_rgb, target_rgb, tolerance=30):
    """判断实际RGB是否匹配目标RGB（允许容差）"""
    if actual_rgb is None or target_rgb is None:
        return False
    return all(abs(actual_rgb[i] - target_rgb[i]) <= tolerance for i in range(3))

def extract_chat_history(screenshot_path, friend_avatar_path, my_avatar_path, device_width, device_height, workflow_folder, region_json=None, friend_rgb=None, my_rgb=None):
    """提取完整的聊天记录"""
    try:
        original_screenshot = read_image_safe(screenshot_path)
        if original_screenshot is None:
            return {'success': False, 'error': '无法读取截图文件'}
        
        # 如果提供了区域，先裁剪图片，然后再进行OCR识别
        # 这样可以确保只识别指定区域，避免识别到键盘和导航栏
        crop_offset_x = 0
        crop_offset_y = 0
        screenshot = original_screenshot
        original_height = original_screenshot.shape[0]
        original_width = original_screenshot.shape[1]
        
        if region_json and region_json != 'None':
            try:
                region = json.loads(region_json)
                # 区域格式：corners 对象，包含 topLeft, topRight, bottomLeft, bottomRight
                if isinstance(region, dict) and 'topLeft' in region and 'bottomRight' in region:
                    top_left = region['topLeft']
                    bottom_right = region['bottomRight']
                    x1 = int(top_left.get('x', 0))
                    y1 = int(top_left.get('y', 0))
                    x2 = int(bottom_right.get('x', original_width))
                    y2 = int(bottom_right.get('y', original_height))
                    
                    # 确保坐标在图片范围内，并且 x2 > x1, y2 > y1
                    x1 = max(0, min(x1, original_width - 1))
                    y1 = max(0, min(y1, original_height - 1))
                    x2 = max(x1 + 1, min(x2, original_width))
                    y2 = max(y1 + 1, min(y2, original_height))
                    
                    # 验证裁剪区域是否有效
                    if x2 > x1 and y2 > y1:
                        # 保存裁剪偏移量（用于调整头像位置）
                        crop_offset_x = x1
                        crop_offset_y = y1
                        # 裁剪图片：使用 numpy 数组切片 [y1:y2, x1:x2]
                        screenshot = original_screenshot[y1:y2, x1:x2]
                        
                        # 验证裁剪后的图片是否有效
                        if screenshot is not None and screenshot.size > 0:
                                # 保存裁剪后的图片到工作流目录下的 tmp 目录，用于调试
                            try:
                                import datetime
                                # 获取工作流目录下的 tmp 目录路径
                                # 方法1: 从 workflow_folder 推断（如果提供）
                                if workflow_folder and workflow_folder != 'None':
                                    workflow_path = Path(workflow_folder)
                                    # workflow_folder 通常是 static/processing/xxx 格式的绝对路径
                                    # tmp 目录应该在工作流目录下：static/processing/xxx/tmp
                                    tmp_dir = workflow_path / 'tmp'
                                else:
                                    # 方法2: 从截图路径推断（向后兼容）
                                    screenshot_path_obj = Path(screenshot_path)
                                    # 尝试向上查找工作流目录（包含 tmp 目录的父目录）
                                    current = screenshot_path_obj.parent
                                    tmp_dir = None
                                    for _ in range(5):  # 最多向上查找5层
                                        if (current / 'tmp').exists():
                                            tmp_dir = current / 'tmp'
                                            break
                                        # 检查是否是工作流目录（包含 processing.json）
                                        if (current / 'processing.json').exists():
                                            tmp_dir = current / 'tmp'
                                            break
                                        current = current.parent
                                    
                                    if tmp_dir is None:
                                        # 如果找不到，使用截图目录的父目录下的 tmp
                                        tmp_dir = screenshot_path_obj.parent / 'tmp'
                                
                                os.makedirs(str(tmp_dir), exist_ok=True)
                                timestamp = datetime.datetime.now().strftime('%Y%m%d_%H%M%S_%f')
                                cropped_image_path = tmp_dir / f'cropped_region_{timestamp}.png'
                                cv2.imwrite(str(cropped_image_path), screenshot)
                                
                                # 清理工作流目录下的 tmp 目录：如果总大小超过 20MB，删除时间最早的文件
                                try:
                                    max_size = 20 * 1024 * 1024  # 20MB
                                    files = []
                                    if tmp_dir.exists():
                                        for file_path in tmp_dir.iterdir():
                                            if file_path.is_file():
                                                file_stat = file_path.stat()
                                                files.append({
                                                    'path': file_path,
                                                    'size': file_stat.st_size,
                                                    'mtime': file_stat.st_mtime
                                                })
                                    
                                    # 计算总大小
                                    total_size = sum(f['size'] for f in files)
                                    
                                    # 如果总大小超过 20MB，删除时间最早的文件
                                    if total_size > max_size:
                                        # 按修改时间排序（最早的在前）
                                        files.sort(key=lambda x: x['mtime'])
                                        
                                        # 删除最早的文件直到总大小小于 20MB
                                        for file_info in files:
                                            if total_size <= max_size:
                                                break
                                            try:
                                                file_info['path'].unlink()
                                                total_size -= file_info['size']
                                            except Exception as del_error:
                                                pass
                                except Exception as cleanup_error:
                                    pass  # 清理失败不影响主流程
                                
                                print(f"Python: 裁剪后的图片已保存到: {cropped_image_path}", file=sys.stderr)
                                print(f"Python: 裁剪区域 - x1={x1}, y1={y1}, x2={x2}, y2={y2}, 原始尺寸={original_width}x{original_height}, 裁剪后尺寸={screenshot.shape[1]}x{screenshot.shape[0]}", file=sys.stderr)
                            except Exception as save_error:
                                print(f"Python: 保存裁剪图片失败: {save_error}", file=sys.stderr)
                                import traceback
                                traceback.print_exc(file=sys.stderr)
                        else:
                            screenshot = original_screenshot
                            crop_offset_x = 0
                            crop_offset_y = 0
            except Exception as e:
                print(f"Python: 区域裁剪异常: {e}", file=sys.stderr)
                import traceback
                traceback.print_exc(file=sys.stderr)
        
        # 查找头像位置
        # 如果图片被裁剪了，在裁剪后的图片上查找头像（需要临时保存）
        # 否则在原始截图上查找
        import tempfile
        temp_cropped_path = None
        try:
            if crop_offset_x > 0 or crop_offset_y > 0:
                # 如果图片被裁剪了，需要临时保存裁剪后的图片用于头像匹配
                temp_cropped_path = tempfile.mktemp(suffix='.png')
                cv2.imwrite(temp_cropped_path, screenshot)
                avatar_positions = find_avatar_positions(temp_cropped_path, friend_avatar_path, my_avatar_path)
            else:
                # 使用原始截图查找头像
                avatar_positions = find_avatar_positions(screenshot_path, friend_avatar_path, my_avatar_path)
        finally:
            # 清理临时文件
            if temp_cropped_path and os.path.exists(temp_cropped_path):
                try:
                    os.remove(temp_cropped_path)
                except:
                    pass
        
        # 获取 OCR 实例
        ocr = ONNXPaddleOcr(use_angle_cls=False, use_gpu=True)
        
        # 执行 OCR（cls=False 避免角度分类器警告）
        # 如果提供了区域，在裁剪后的图片上识别；否则全屏识别
        ocr_result = ocr.ocr(screenshot, cls=False)
        
        if not ocr_result or not ocr_result[0]:
            return {'success': False, 'error': 'OCR 识别失败'}
        
        # 解析 OCR 结果，按 y 坐标分组消息
        messages = []
        friend_positions = avatar_positions.get('friend', [])
        my_positions = avatar_positions.get('my', [])
        
        # 获取截图高度（如果被裁剪了，使用裁剪后的高度；否则使用原始高度）
        screenshot_height = screenshot.shape[0]
        screenshot_width = screenshot.shape[1]
        
        # 计算键盘区域的阈值：通常键盘在屏幕底部，占屏幕高度的30-40%
        # 如果提供了区域裁剪，说明用户已经指定了识别区域，应该信任这个区域，不进行键盘过滤
        # 或者使用更宽松的阈值（90%），只过滤最底部的内容
        if region_json and region_json != 'None':
            # 如果提供了识别区域，使用更宽松的阈值（90%），几乎不过滤
            # 因为用户已经通过区域限制了识别范围
            keyboard_threshold_y = int(screenshot_height * 0.90)  # 从90%的位置开始过滤
        else:
            # 如果没有提供区域，使用原来的65%阈值
            keyboard_threshold_y = int(screenshot_height * 0.65)  # 从65%的位置开始过滤
        
        # 简单的消息分组逻辑：根据 y 坐标和头像位置判断发送者
        for line in ocr_result[0]:
            if not line:
                continue
            
            box = line[0]
            text = line[1][0] if len(line) > 1 and line[1] else ''
            confidence = line[1][1] if len(line) > 1 and len(line[1]) > 1 else 0.0
            
            if not text or confidence < 0.5:
                continue
            
            # 计算消息框的中心坐标
            x_center = sum([point[0] for point in box]) / len(box)
            y_center = sum([point[1] for point in box]) / len(box)
            
            # 过滤键盘区域：如果y坐标超过阈值，直接跳过
            if y_center > keyboard_threshold_y:
                continue
            
            # 过滤明显是键盘按键的文本（即使y坐标在阈值内，也要过滤）
            keyboard_keywords = ['ABC', 'DEF', 'GHI', 'JKL', 'MNO', 'PQRS', 'TUV', 'WXYZ', 
                               '分词', '重输', '换行', '符', '中/英', '中二', '123', '0', '1', '2', '3', 
                               '4', '5', '6', '7', '8', '9', 'Q', 'W', 'E', 'R', 'T', 'Y', 'U', 'I', 
                               'O', 'P', 'A', 'S', 'D', 'F', 'G', 'H', 'J', 'K', 'L', 'Z', 'X', 'C', 
                               'V', 'B', 'N', 'M', '△', '三', '-']
            if text.strip() in keyboard_keywords:
                continue
            
            # 判断发送者（优先使用RGB颜色判断，如果失败则使用头像距离判断，最后使用 x 坐标判断）
            sender = 'unknown'
            
            # 方法1: 使用RGB颜色判断（最高优先级）
            if friend_rgb and my_rgb:
                try:
                    bubble_rgb = detect_bubble_color(screenshot, box)
                    if bubble_rgb:
                        # 判断是否匹配好友颜色（白色/浅色）
                        if match_rgb_color(bubble_rgb, friend_rgb, tolerance=40):
                            sender = 'friend'
                        # 判断是否匹配我的颜色（绿色）
                        elif match_rgb_color(bubble_rgb, my_rgb, tolerance=40):
                            sender = 'me'
                except Exception as e:
                    pass  # RGB检测失败，继续使用其他方法
            
            # 方法2: 如果RGB判断失败，使用头像位置距离判断
            if sender == 'unknown':
                min_friend_dist = float('inf')
                min_my_dist = float('inf')
                
                for fx, fy in friend_positions:
                    dist = abs(y_center - (fy + 20))  # 假设头像高度约 40px
                    if dist < min_friend_dist:
                        min_friend_dist = dist
                
                for mx, my in my_positions:
                    dist = abs(y_center - (my + 20))
                    if dist < min_my_dist:
                        min_my_dist = dist
                
                # 优先使用距离判断（阈值100像素）
                if min_friend_dist < 100 and min_friend_dist < min_my_dist:
                    sender = 'friend'
                elif min_my_dist < 100 and min_my_dist < min_friend_dist:
                    sender = 'me'
                else:
                    # 距离判断失败，使用备选方法
                    screen_center_x = device_width / 2
                    # 如果两个头像都没找到，直接使用 x 坐标判断
                    if not friend_positions and not my_positions:
                        if x_center < screen_center_x:
                            sender = 'friend'  # 左侧通常是好友
                        else:
                            sender = 'me'  # 右侧通常是"我"
                    # 如果只找到好友头像，放宽阈值到150像素
                    elif friend_positions and not my_positions:
                        if min_friend_dist < 150:
                            sender = 'friend'
                        elif x_center < screen_center_x:
                            sender = 'friend'
                        else:
                            sender = 'me'
                    # 如果只找到我的头像，放宽阈值到150像素
                    elif my_positions and not friend_positions:
                        if min_my_dist < 150:
                            sender = 'me'
                        elif x_center >= screen_center_x:
                            sender = 'me'
                        else:
                            sender = 'friend'
                    # 如果两个头像都找到了但距离判断失败，使用 x 坐标判断
                    else:
                        if x_center < screen_center_x:
                            sender = 'friend'
                        else:
                            sender = 'me'
            
            messages.append({
                'text': text,
                'sender': sender,
                'y': int(y_center),
                'confidence': float(confidence)
            })
        
        # 按 y 坐标排序（从上到下）
        messages.sort(key=lambda m: m['y'])
        
        # 格式化消息文本
        messages_text = '\\n'.join([f"{'对方' if m['sender'] == 'friend' else '我' if m['sender'] == 'me' else '未知'}: {m['text']}" for m in messages])
        
        result = {
            'success': True,
            'messages': messages,
            'messagesText': messages_text,
            'count': len(messages)
        }
        
        return result
        
    except Exception as e:
        return {'success': False, 'error': f'提取聊天记录失败: {str(e)}'}

# 主逻辑
if __name__ == '__main__':
    import sys
    screenshot_path = sys.argv[1]
    friend_avatar_path = sys.argv[2] if len(sys.argv) > 2 and sys.argv[2] != 'None' else None
    my_avatar_path = sys.argv[3] if len(sys.argv) > 3 and sys.argv[3] != 'None' else None
    device_width = int(sys.argv[4]) if len(sys.argv) > 4 and sys.argv[4] else 1080
    device_height = int(sys.argv[5]) if len(sys.argv) > 5 and sys.argv[5] else 2400
    workflow_folder = sys.argv[6] if len(sys.argv) > 6 and sys.argv[6] != 'None' else None
    region_json = sys.argv[7] if len(sys.argv) > 7 and sys.argv[7] != 'None' else None
    friend_rgb_str = sys.argv[8] if len(sys.argv) > 8 and sys.argv[8] != 'None' else None
    my_rgb_str = sys.argv[9] if len(sys.argv) > 9 and sys.argv[9] != 'None' else None
    
    # 解析RGB字符串（格式："(r,g,b)"）
    friend_rgb = None
    my_rgb = None
    
    if friend_rgb_str and friend_rgb_str != 'None':
        try:
            # 解析格式 "(r,g,b)" 或 "(r, g, b)"
            friend_rgb_str = friend_rgb_str.strip().strip('()')
            parts = [int(x.strip()) for x in friend_rgb_str.split(',')]
            if len(parts) == 3:
                friend_rgb = parts
        except Exception as e:
            print(f"Python: 解析好友RGB失败: {e}", file=sys.stderr)
    
    if my_rgb_str and my_rgb_str != 'None':
        try:
            # 解析格式 "(r,g,b)" 或 "(r, g, b)"
            my_rgb_str = my_rgb_str.strip().strip('()')
            parts = [int(x.strip()) for x in my_rgb_str.split(',')]
            if len(parts) == 3:
                my_rgb = parts
        except Exception as e:
            print(f"Python: 解析我的RGB失败: {e}", file=sys.stderr)
    
    # 打印接收到的参数（用于调试）
    print(f"Python: 接收到的参数 - region_json={'已提供' if region_json and region_json != 'None' else '未提供'}, friend_rgb={friend_rgb}, my_rgb={my_rgb}", file=sys.stderr)
    
    result = extract_chat_history(screenshot_path, friend_avatar_path, my_avatar_path, device_width, device_height, workflow_folder, region_json, friend_rgb, my_rgb)
    print(json.dumps(result, ensure_ascii=False))
`;

    // 将 Python 代码写入临时文件
    const tempScriptPath = join(__dirname, '..', '..', 'temp_extract_chat_history.py');
    await writeFile(tempScriptPath, pythonCode, 'utf8');

    // 构建命令
    const normalizedScreenshotPath = screenshotPath.replace(/\\/g, '/');
    let friendAvatarArg = 'None';
    if (friendAvatarPath) {
      friendAvatarArg = isAbsolute(friendAvatarPath) 
        ? friendAvatarPath.replace(/\\/g, '/')
        : join(__dirname, '..', '..', 'static', 'processing', friendAvatarPath).replace(/\\/g, '/');
    }
    
    let myAvatarArg = 'None';
    if (myAvatarPath) {
      myAvatarArg = isAbsolute(myAvatarPath)
        ? myAvatarPath.replace(/\\/g, '/')
        : join(__dirname, '..', '..', 'static', 'processing', myAvatarPath).replace(/\\/g, '/');
    }
    
    let workflowFolderArg = 'None';
    if (workflowFolder) {
      workflowFolderArg = isAbsolute(workflowFolder)
        ? workflowFolder.replace(/\\/g, '/')
        : join(__dirname, '..', '..', 'static', 'processing', workflowFolder).replace(/\\/g, '/');
    }

    // 传递区域参数（如果提供）
    let regionArg = 'None';
    if (regionJson && regionJson !== 'None') {
      regionArg = regionJson.replace(/"/g, '\\"');
    }
    
    // 传递RGB参数（如果提供）
    let friendRgbArg = 'None';
    if (friendRgb && typeof friendRgb === 'string') {
      friendRgbArg = friendRgb;
    }
    
    let myRgbArg = 'None';
    if (myRgb && typeof myRgb === 'string') {
      myRgbArg = myRgb;
    }
    
    const command = `"${pythonExePath}" "${tempScriptPath}" "${normalizedScreenshotPath}" "${friendAvatarArg}" "${myAvatarArg}" ${deviceWidth || 1080} ${deviceHeight || 2400} "${workflowFolderArg}" "${regionArg}" "${friendRgbArg}" "${myRgbArg}"`;

    const env = {
      ...process.env,
      DISABLE_MODEL_SOURCE_CHECK: 'True'
    };

    const { stdout, stderr } = await execAsync(command, {
      timeout: 60000,
      maxBuffer: 10 * 1024 * 1024,
      cwd: join(__dirname, '..', '..'),
      encoding: 'utf8',
      env: { ...env, PYTHONIOENCODING: 'utf-8', PYTHONUTF8: '1' }
    });

    // 打印 Python 脚本的 stderr 输出
    if (stderr && stderr.trim()) {
      try {
        const decodedStderr = Buffer.from(stderr, 'utf8').toString('utf8');
        console.log(decodedStderr.trim());
      } catch (e) {
        console.log(stderr.trim());
      }
    }

    // 清理临时文件
    try {
      await import('fs/promises').then(fs => fs.unlink(tempScriptPath));
    } catch (e) {
      // 忽略删除失败
    }

    // 解析输出
    const cleanStdout = stdout.replace(/\[33m.*?\[0m/g, '').replace(/DeprecationWarning.*?\n/g, '');
    
    try {
      const result = JSON.parse(cleanStdout.trim());
      return result;
    } catch (parseError) {
      console.error('聊天记录解析失败:', parseError);
      console.error('原始输出:', cleanStdout);
      return { success: false, error: `解析聊天记录失败: ${parseError.message}` };
    }
  } catch (error) {
    console.error('提取聊天记录失败:', error);
    if (error.message && error.message.includes('timeout')) {
      return { success: false, error: '提取聊天记录超时，请检查网络连接或稍后重试' };
    }
    return { success: false, error: error.message };
  }
}

/**
 * 获取最后一条消息
 */
export async function getLastMessage(screenshotPath, friendAvatarPath, myAvatarPath, deviceWidth, deviceHeight) {
  try {
    // 先提取完整聊天记录
    const result = await extractChatHistory(screenshotPath, friendAvatarPath, myAvatarPath, deviceWidth, deviceHeight, null);
    
    if (!result.success || !result.messages || result.messages.length === 0) {
      return { success: false, error: '未找到消息' };
    }
    
    // 获取最后一条消息（y 坐标最大的）
    const lastMessage = result.messages.reduce((max, msg) => msg.y > max.y ? msg : max, result.messages[0]);
    
    return {
      success: true,
      text: lastMessage.text,
      sender: lastMessage.sender,
      position: { y: lastMessage.y }
    };
  } catch (error) {
    return { success: false, error: error.message };
  }
}

/**
 * 全屏 OCR 识别
 */
export async function ocrFullScreen(screenshotPath, deviceWidth, deviceHeight) {
  try {
    const pythonExePath = join(__dirname, '..', '..', 'py', 'venv', 'Scripts', 'python.exe');
    const onnxocrPath = join(__dirname, '..', '..', 'py', 'OnnxOCR');
    
    const pythonCode = `
import sys
import os
import cv2
import numpy as np
import json

# 添加 OnnxOCR 路径
sys.path.insert(0, r"${onnxocrPath.replace(/\\/g, '/')}")

from onnxocr.onnx_paddleocr import ONNXPaddleOcr

# 设置环境变量
os.environ['DISABLE_MODEL_SOURCE_CHECK'] = 'True'

def read_image_safe(image_path):
    abs_path = os.path.abspath(str(image_path))
    try:
        with open(abs_path, 'rb') as f:
            image_data = f.read()
        img_array = np.frombuffer(image_data, np.uint8)
        img = cv2.imdecode(img_array, cv2.IMREAD_COLOR)
        if img is None:
            raise ValueError(f"cv2.imdecode 无法解码图片: {abs_path}")
        return img
    except FileNotFoundError:
        raise FileNotFoundError(f"图片文件不存在: {abs_path}")
    except Exception as e:
        raise Exception(f"读取图片失败: {abs_path}, 错误: {str(e)}")

# 主逻辑
if __name__ == '__main__':
    screenshot_path = sys.argv[1]
    
    try:
        screenshot = read_image_safe(screenshot_path)
        if screenshot is None:
            print(json.dumps({'success': False, 'error': '无法读取截图文件'}, ensure_ascii=False))
            sys.exit(1)
        
        ocr = ONNXPaddleOcr(use_angle_cls=False, use_gpu=True)
        ocr_result = ocr.ocr(screenshot, cls=False)
        
        if not ocr_result or not ocr_result[0]:
            print(json.dumps({'success': False, 'error': 'OCR 识别失败'}, ensure_ascii=False))
            sys.exit(1)
        
        # 提取所有文本
        texts = []
        for line in ocr_result[0]:
            if line and len(line) > 1:
                text = line[1][0] if isinstance(line[1], (list, tuple)) else str(line[1])
                if text:
                    texts.append(text)
        
        full_text = '\\n'.join(texts)
        
        print(json.dumps({
            'success': True,
            'text': full_text,
            'position': None
        }, ensure_ascii=False))
    except Exception as e:
        print(json.dumps({'success': False, 'error': f'OCR 识别失败: {str(e)}'}, ensure_ascii=False))
        sys.exit(1)
`;

    // 将 Python 代码写入临时文件
    const tempScriptPath = join(__dirname, '..', '..', 'temp_ocr_full_screen.py');
    await writeFile(tempScriptPath, pythonCode, 'utf8');

    const normalizedScreenshotPath = screenshotPath.replace(/\\/g, '/');
    const command = `"${pythonExePath}" "${tempScriptPath}" "${normalizedScreenshotPath}"`;

    const env = {
      ...process.env,
      DISABLE_MODEL_SOURCE_CHECK: 'True'
    };

    const { stdout, stderr } = await execAsync(command, {
      timeout: 60000,
      maxBuffer: 10 * 1024 * 1024,
      cwd: join(__dirname, '..', '..'),
      encoding: 'utf8',
      env: { ...env, PYTHONIOENCODING: 'utf-8', PYTHONUTF8: '1' }
    });

    // 清理临时文件
    try {
      await import('fs/promises').then(fs => fs.unlink(tempScriptPath));
    } catch (e) {
      // 忽略删除失败
    }

    const cleanStdout = stdout.replace(/\[33m.*?\[0m/g, '').replace(/DeprecationWarning.*?\n/g, '');

    try {
      const result = JSON.parse(cleanStdout.trim());
      return result;
    } catch (parseError) {
      return { success: false, error: `解析失败: ${parseError.message}` };
    }
  } catch (error) {
    return { success: false, error: error.message };
  }
}