智能文档处理平台的全面测评：Textin深度对比与API调用

self._app_id = 'c81f*************************e9ff'
# 请登录后前往 “工作台-账号设置-开发者信息” 查看 x-ti-secret-code
# 示例代码中 x-ti-secret-code 非真实数据
self._secret_code = '5508***********************1c17'

if __name__ == "__main__":
    # 示例 1：传输文件
    response = CommonOcr(img_path=r'example.jpg')
    print(response.recognize())
    # 示例 2：传输 URL
    response = CommonOcr(img_path='http://example.com/example.jpg', is_url=True)
    print(response.recognize())

import requests
import json
from typing import List, Dict

def get_file_content(filePath):
    with open(filePath, 'rb') as fp:
        return fp.read()

class CommonOcr(object):
    def __init__(self, img_path=None, is_url=False):
        self._url = 'https://api.textin.com/ai/service/v2/recognize/multipage'
        self._app_id = '25a843d9f236d9ee743443897549f856'  # 替换为真实ID
        self._secret_code = '183f28d1c9e447b08e50c19fd22504e3'  # 替换为真实密钥
        self._img_path = img_path
        self._is_url = is_url

    def recognize(self) -> Dict:
        """调用API并返回解析后的字典"""
        head = {
            'x-ti-app-id': self._app_id,
            'x-ti-secret-code': self._secret_code
        }
        try:
            if self._is_url:
                head['Content-Type'] = 'text/plain'
                body = self._img_path
            else:
                image = get_file_content(self._img_path)
                head['Content-Type'] = 'application/octet-stream'
                body = image
            
            result = requests.post(self._url, data=body, headers=head)
            return json.loads(result.text)
        except Exception as e:
            return {"error": str(e)}

def format_recognition_result(
    result: Dict, 
    spacing_threshold: float = 1.2,  # 降低分段阈值（适合紧凑排版）
    force_break_words: List[str] = ["分", "章", "節"]  # 遇到这些词强制分段（适配古籍）
) -> str:
    """优化分段逻辑：支持阈值调整+关键词强制分段"""
    if "error" in result:
        return f"识别错误：{result['error']}"
    
    if result.get("code") != 200:
        return f"API调用失败：{result.get('msg', '未知错误')}"
    
    pages = result.get("result", {}).get("pages", [])
    if not pages:
        return "未识别到任何文字"
    
    lines: List[Dict] = pages[0].get("lines", [])
    if not lines:
        return "未识别到文字内容"
    
    # 判断排版方向
    is_vertical = lines[0].get("direction", 0) == 2
    
    # 排序（按阅读顺序）
    if is_vertical:
        sorted_lines = sorted(lines, key=lambda x: (-x["position"][0], x["position"][1]))
    else:
        sorted_lines = sorted(lines, key=lambda x: (x["position"][1], x["position"][0]))
    
    # 计算每行高度（用于分段判断）
    line_heights = []
    for line in sorted_lines:
        y_coords = [line["position"][1], line["position"][3], line["position"][5], line["position"][7]]
        line_heights.append(max(y_coords) - min(y_coords))
    avg_line_height = sum(line_heights) / len(line_heights) if line_heights else 10
    
    # 分段逻辑：结合间距和关键词
    paragraphs = []
    current_paragraph = []
    
    for i, line in enumerate(sorted_lines):
        line_text = line["text"]
        current_paragraph.append(line_text)
        
        # 强制分段：如果当前行包含指定关键词，且不是最后一行
        force_break = any(word in line_text for word in force_break_words)
        if force_break and i < len(sorted_lines) - 1:
            paragraphs.append("".join(current_paragraph))
            current_paragraph = []
            continue
        
        # 按间距分段（最后一行不需要判断）
        if i < len(sorted_lines) - 1:
            next_line = sorted_lines[i+1]
            # 计算当前行与下一行的间距
            if is_vertical:
                # 竖排：当前行底部y坐标 vs 下一行顶部y坐标
                current_bottom = max([line["position"][1], line["position"][3], line["position"][5], line["position"][7]])
                next_top = min([next_line["position"][1], next_line["position"][3], next_line["position"][5], next_line["position"][7]])
                spacing = next_top - current_bottom
            else:
                # 横排：当前行右侧x坐标 vs 下一行左侧x坐标
                current_right = max([line["position"][0], line["position"][2], line["position"][4], line["position"][6]])
                next_left = min([next_line["position"][0], next_line["position"][2], next_line["position"][4], next_line["position"][6]])
                spacing = next_left - current_right
            
            # 当间距超过平均行高的N倍时分段
            if spacing > avg_line_height * spacing_threshold:
                paragraphs.append("".join(current_paragraph))
                current_paragraph = []
    
    # 加入最后一段
    if current_paragraph:
        paragraphs.append("".join(current_paragraph))
    
    return "\n\n".join(paragraphs)

if __name__ == "__main__":
    # 调用API
    ocr = CommonOcr(img_path=r'C:\Users\ZHANGZHANHUA\PycharmProjects\pythonProject\.venv\example.png')
    result = ocr.recognize()
    
    # 格式化（降低阈值+古籍关键词分段）
    formatted_text = format_recognition_result(
        result,
        spacing_threshold=1.2,  # 阈值降低，更容易分段
        force_break_words=["分", "經", "佛", "爾時"]  # 遇到这些词强制分段（根据金刚经特点调整）
    )
    
    # 打印结果
    print("识别结果：")
    print("=" * 50)
    print(formatted_text)
    print("=" * 50)
    
    # 统计行数
    try:
        pages = result.get('result', {}).get('pages', [])
        line_count = len(pages[0].get('lines', [])) if pages else 0
        print(f"识别完成（共{line_count}行文字）")
    except:
        print("无法统计行数")