(\ _ /)
( ･-･)
/っ 
就是第 2 个东西，第 3 个不支持图片识别就不折腾了。
价格不重要，有赠金的时候才折腾，没赠金就嫖别的渠道.jpg
总而言之鼠标点点创建就行，会给一个 API 接口。
然后直接上源码：
DocumentaiFormExtractor.py
# pip install google-cloud-documentai
# pip install grpcio==1.50.0 # Windows 7 只能兼容到这个版本。
# pip install tqdm # 用于进度条显示
# 导入必要的库
from google.api_core.client_options import ClientOptions
from google.cloud import documentai # type: ignore
import tkinter as tk
from tkinter import filedialog
import os
import threading
from concurrent.futures import ThreadPoolExecutor, as_completed
from tqdm import tqdm
import time
# 设置代理
os.environ.update({'http_proxy': 'http://127.0.0.1:1081', 'https_proxy': 'http://127.0.0.1:1081'})
# 设置服务帐户密钥文件路径
os.environ["GOOGLE_APPLICATION_CREDENTIALS"] = "(你的鉴权文件).json"
# 配置信息
project_id = "(这里改成你端点的)" # 你的 GCP 项目 ID
location = "us" # 处理器位置，可以是 "us" 或 "eu"
processor_id = "(这里改成你端点的)" # 你的 Form Parser 处理器ID
max_workers = 5 # 最大并行线程数，可根据需要调整
def process_single_document(file_path, client, processor_name):
"""处理单个文档的函数"""
try:
# 读取文件内容到内存
with open(file_path, "rb") as image:
image_content = image.read()
# 根据文件扩展名确定 mime_type
file_extension = os.path.splitext(file_path)[1].lower()
if file_extension == ".pdf":
mime_type = "application/pdf"
elif file_extension in [".jpg", ".jpeg"]:
mime_type = "image/jpeg"
elif file_extension == ".png":
mime_type = "image/png"
elif file_extension == ".gif":
mime_type = "image/gif"
elif file_extension in [".tif", ".tiff"]:
mime_type = "image/tiff"
elif file_extension == ".bmp":
mime_type = "image/bmp"
else:
raise ValueError(f"不支持的文件格式: {file_extension}")
# 加载二进制数据
raw_document = documentai.RawDocument(
content=image_content,
mime_type=mime_type,
)
# 配置处理请求
request = documentai.ProcessRequest(name=processor_name, raw_document=raw_document)
# 调用 Document AI 服务处理文档
result = client.process_document(request=request)
document = result.document
# 获取所选文件的目录和文件名（不含扩展名）
file_dir = os.path.dirname(file_path)
file_name = os.path.splitext(os.path.basename(file_path))[0]
# 构建输出文件的完整路径（JSON 格式）
output_file_path = os.path.join(file_dir, f"{file_name}.json")
# 将识别的表单数据写入到输出文件（JSON 格式）
with open(output_file_path, "w", encoding="utf-8") as output_file:
output_file.write(documentai.Document.to_json(document)) # 使用 to_json 方法
return True, file_path, output_file_path
except Exception as e:
return False, file_path, str(e)
def batch_process_documents():
"""批量处理文档的主函数"""
# 初始化 Document AI 客户端
opts = ClientOptions(api_endpoint=f"{location}-documentai.googleapis.com")
client = documentai.DocumentProcessorServiceClient(client_options=opts)
processor_name = f"projects/{project_id}/locations/{location}/processors/{processor_id}"
# 创建文件选择对话框
root = tk.Tk()
root.withdraw() # 隐藏主窗口
# 弹出文件选择对话框，允许多选, 并添加更多文件类型
file_paths = filedialog.askopenfilenames(
title="选择多个文件",
filetypes=[
("所有支持的格式", "*.pdf;*.jpg;*.jpeg;*.png;*.gif;*.tif;*.tiff;*.bmp"),
("PDF 文件", "*.pdf"),
("JPEG 图片", "*.jpg;*.jpeg"),
("PNG 图片", "*.png"),
("GIF 图片", "*.gif"),
("TIFF 图片", "*.tif;*.tiff"),
("BMP 图片", "*.bmp"),
]
)
# 如果用户取消了文件选择，则退出程序
if not file_paths or len(file_paths) == 0:
print("未选择文件，程序退出。")
return
total_files = len(file_paths)
print(f"已选择 {total_files} 个文件，开始处理...")
# 创建进度条
progress_bar = tqdm(total=total_files, desc="处理进度", unit="文件")
# 使用线程池并行处理文件
results = []
with ThreadPoolExecutor(max_workers=max_workers) as executor:
# 提交所有任务
future_to_file = {
executor.submit(process_single_document, file_path, client, processor_name): file_path
for file_path in file_paths
}
# 处理完成的任务
for future in as_completed(future_to_file):
file_path = future_to_file[future]
try:
success, path, result = future.result()
if success:
results.append((path, result, True))
else:
results.append((path, result, False))
except Exception as e:
results.append((file_path, str(e), False))
# 更新进度条
progress_bar.update(1)
# 关闭进度条
progress_bar.close()
# 显示处理结果统计
success_count = sum(1 for _, _, success in results if success)
failed_count = total_files - success_count
print(f"\n处理完成！成功: {success_count} 个文件，失败: {failed_count} 个文件")
# 如果有失败的文件，显示详情
if failed_count > 0:
print("\n失败文件列表:")
for path, error, success in results:
if not success:
print(f"- {os.path.basename(path)}: {error}")
# 显示所有成功处理的文件路径
if success_count > 0:
print("\n成功处理的文件:")
for path, output, success in results:
if success:
print(f"- {os.path.basename(path)} -> {os.path.basename(output)}")
if __name__ == "__main__":
batch_process_documents()
这个跑完会返回一个 json，因为包含了原图 base64 所以比较巨大。
然后这个是转成 html 预览的脚本 
DocumentaiFormJsonViewer.py
import json
import tkinter as tk
from tkinter import filedialog
import os
from google.cloud import documentai_v1 as documentai
from PIL import Image
import io
import base64
def get_bounding_box_str(entity):
"""Helper function to format bounding box vertices."""
try:
if entity.bounding_poly and entity.bounding_poly.normalized_vertices:
vertices = entity.bounding_poly.normalized_vertices
return ", ".join([f"({v.x:.3f}, {v.y:.3f})" for v in vertices])
else:
return "N/A"
except AttributeError:
return "N/A"
def convert_json_to_html(json_file_path, font_size=12, image_opacity=0.5, show_visual_elements=False):
"""
Converts Document AI JSON to HTML with layout preservation.
Args:
json_file_path: Path to the Document AI JSON file.
font_size: The font size (in pixels) to use for the text.
image_opacity: The opacity of the background image (0.0 to 1.0).
show_visual_elements: Whether to show visual elements (like checkboxes).
"""
try:
with open(json_file_path, 'rb') as f:
json_data = f.read()
document = documentai.Document.from_json(json_data)
except Exception as e:
print(f"Error loading document: {e}")
return
html_output = f"""
<!DOCTYPE html>
<html><head>
<title>Document AI Output</title>
<style>
.page-container {{ position: relative; border: 1px solid black; }}
/* 动态宽度，不截断文本 */
.positioned-element {{ position: absolute; font-size: {font_size}px; }}
.image-element {{ position: absolute; opacity: {image_opacity}; }}
.checkbox {{ position: absolute; width: 12px; height: 12px; border: 1px solid gray; }}
.checked {{ background-color: black; }}
/* 视觉元素样式 */
.visual-element {{ position: absolute; border: 1px dashed blue; width: 12px; height: 12px;}}
.bbox {{ color: blue; font-size: smaller; }}
</style>
</head><body>
"""
for page_number, page in enumerate(document.pages):
width = int(page.dimension.width)
height = int(page.dimension.height)
html_output += f'<div class="page-container" style="width: {width}px; height: {height}px;">'
# 1. 图像 (放在最底层)
if page.image:
try:
image_data = page.image.content
image = Image.open(io.BytesIO(image_data))
buffered = io.BytesIO()
image.save(buffered, format="PNG")
img_str = base64.b64encode(buffered.getvalue()).decode()
html_output += f'<img class="image-element" src="data:image/png;base64,{img_str}" style="left: 0px; top: 0px; width: {width}px; height: {height}px; z-index: 1; pointer-events: none;" disabled>'
except Exception as e:
print(f"Error processing image: {e}")
# 2. 文本块
for block in page.blocks:
bbox = block.layout.bounding_poly.normalized_vertices
if bbox:
x1, y1 = int(bbox[0].x * width), int(bbox[0].y * height)
x2, y2 = int(bbox[2].x * width), int(bbox[2].y * height)
text = get_text(block.layout, document)
html_output += f'<div class="positioned-element" style="left: {x1}px; top: {y1}px; z-index: 3;">{text}</div>'
# 3. 表单字段
for field in page.form_fields:
# 字段名
if field.field_name:
name_bbox = field.field_name.bounding_poly.normalized_vertices
name_text = get_text(field.field_name, document) or ""
if name_bbox:
x1, y1 = int(name_bbox[0].x * width), int(name_bbox[0].y * height)
x2, y2 = int(name_bbox[2].x * width), int(name_bbox[2].y * height)
html_output += f'<div class="positioned-element" style="left: {x1}px; top: {y1}px; font-weight: bold; z-index: 3;">{name_text.strip()}</div>'
# 字段值
if field.field_value:
value_bbox = field.field_value.bounding_poly.normalized_vertices
value_text = get_text(field.field_value, document) or ""
if value_bbox:
x1, y1 = int(value_bbox[0].x * width), int(value_bbox[0].y * height)
x2, y2 = int(value_bbox[2].x * width), int(value_bbox[2].y * height)
html_output += f'<div class="positioned-element" style="left: {x1}px; top: {y1}px; z-index: 3;">{value_text.strip()}</div>'
# 4. 视觉元素 (可选显示)
if show_visual_elements:
for visual_element in page.visual_elements:
bbox = visual_element.layout.bounding_poly.normalized_vertices
if bbox:
x1, y1 = int(bbox[0].x * width), int(bbox[0].y * height)
x2, y2 = int(bbox[2].x * width), int(bbox[2].y * height)
# 为所有视觉元素添加通用样式, 并设置较小的尺寸
html_output += f'<div class="visual-element" style="left: {x1}px; top: {y1}px; z-index: 2;"></div>'
html_output += "</div>"
html_output += "</body></html>"
base_name = os.path.splitext(json_file_path)[0]
html_file_path = base_name + ".html"
with open(html_file_path, "w", encoding="utf-8") as f:
f.write(html_output)
print(f"HTML output saved to: {html_file_path}")
def get_text(entity, document):
"""Extracts text from an entity, handling cases where text_anchor might be None."""
text = ""
if hasattr(entity, 'text_anchor') and entity.text_anchor:
if entity.text_anchor.text_segments:
for segment in entity.text_anchor.text_segments:
start_index = int(segment.start_index)
end_index = int(segment.end_index)
text += document.text[start_index:end_index]
return text
def open_file_dialog():
root = tk.Tk()
root.withdraw()
file_path = filedialog.askopenfilename(title="Select Document AI JSON File", filetypes=[("JSON files", "*.json")])
if file_path:
# 示例：设置字体大小、图像透明度。不显示视觉元素
convert_json_to_html(file_path, font_size=18, image_opacity=0.1, show_visual_elements=False)
if __name__ == "__main__":
open_file_dialog()
(\ _ /)
( ･-･)
/っ 
就是这样。效果
感谢佬友分享