Echo/fcb_photo_review
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

优化图片的存储，及时删除处理过程中产生的图片

Browse Source
This commit is contained in:
liuyebo
2024-10-15 10:17:01 +08:00
parent 15ea3ff96f
commit 5af6256376
5 changed files with 141 additions and 94 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
photo_mask/auto_photo_mask.py
View File

@@ -54,11 +54,11 @@ def find_boxes(content, layout, offset=0, length=None, improve=False, image_path
if improve:
# 再次识别，提高精度
image = cv2.imread(image_path)
img_name, img_ext = image_util.parse_save_path(image_path)
img_name, img_ext = common_util.parse_save_path(image_path)
# 截图时偏大一点
capture_box = common_util.zoom_rectangle(box, 0.2)
captured_image = image_util.capture(image, capture_box)
captured_image_path = image_util.get_save_path(f'{img_name}.capture.{img_ext}')
captured_image_path = common_util.get_processed_img_path(f'{img_name}.capture.{img_ext}')
cv2.imwrite(captured_image_path, captured_image)
captured_a4_img_path, offset_x, offset_y = image_util.expand_to_a4_size(captured_image_path)
try:
@@ -170,7 +170,7 @@ def handle_image_for_mask(split_result):
def mask_photo(img_path, name, id_card_num, color=(255, 255, 255)):
def _mask(ip, n, icn, c):
i = cv2.imread(ip)
img_name, img_ext = image_util.parse_save_path(ip)
img_name, img_ext = common_util.parse_save_path(ip)
do_mask = False
split_results = image_util.split(ip)
for split_result in split_results:
@@ -188,7 +188,7 @@ def mask_photo(img_path, name, id_card_num, color=(255, 255, 255)):
result[3] + y_offset,
)
cv2.rectangle(i, (int(result[0]), int(result[1])), (int(result[2]), int(result[3])), c, -1, 0)
masked_path = image_util.get_save_path(f'{img_name}.mask.{img_ext}')
masked_path = common_util.get_processed_img_path(f'{img_name}.mask.{img_ext}')
cv2.imwrite(masked_path, i)
return do_mask, masked_path
@@ -222,7 +222,7 @@ def photo_mask(pk_phhd, name, id_card_num):
img_url = ufile.get_private_url(phrec.cfjaddress)
if not img_url:
continue
img_path = image_util.save_to_local(img_url)
img_path = common_util.save_to_local(img_url)
is_masked, image = mask_photo(img_path, name, id_card_num)
# 如果涂抹了要备份以及更新

8
photo_review/__init__.py
View File

@@ -9,6 +9,8 @@ PHHD_BATCH_SIZE = 10
SLEEP_MINUTES = 5
# 是否发送报错邮件
SEND_ERROR_EMAIL = True
# 处理批号（这里仅起声明作用）
BATCH_ID = ''
'''
别名配置
@@ -56,3 +58,9 @@ DISCHARGE_KEY = {
'出院日期': ['出院日期', '出院时间'],
'出院医嘱': ['出院医嘱']
}
def modify_batch_id(batch_id):
# 修改批号的接口
global BATCH_ID
BATCH_ID = batch_id

68
photo_review/auto_photo_review.py
View File

@@ -1,6 +1,9 @@
import json
import logging
import os
import shutil
import time
import uuid
from collections import defaultdict
from time import sleep
@@ -16,7 +19,7 @@ from db import MysqlSession
from db.mysql import BdYljg, BdYlks, ZxIeCost, ZxIeDischarge, ZxIeSettlement, ZxPhhd, ZxPhrec, ZxIeReview, ZxIeResult
from log import HOSTNAME
from photo_review import PHHD_BATCH_SIZE, SLEEP_MINUTES, HOSPITAL_ALIAS, HOSPITAL_FILTER, DEPARTMENT_ALIAS, \
DEPARTMENT_FILTER, DISCHARGE_KEY
DEPARTMENT_FILTER, DISCHARGE_KEY, modify_batch_id, BATCH_ID
from services.paddle_services import IE_KEY
from ucloud import ufile
from util import image_util, common_util, html_util, model_util
@@ -46,8 +49,8 @@ def get_better_image_from_qrcode(img_path, image_id, dpi=150):
# 将渲染结果转换为OpenCV兼容的格式
img = np.frombuffer(pix.samples, dtype=np.uint8).reshape((pix.height, pix.width, -1))
img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
img_name, img_ext = image_util.parse_save_path(img_path)
better_img_path = image_util.get_save_path(f'{img_name}.better.{img_ext}')
img_name, img_ext = common_util.parse_save_path(img_path)
better_img_path = common_util.get_processed_img_path(f'{img_name}.better.{img_ext}')
cv2.imwrite(better_img_path, img)
return better_img_path, page.get_text()
except Exception as ex:
@@ -100,20 +103,21 @@ def get_better_image_from_qrcode(img_path, image_id, dpi=150):
# 关键信息提取
def information_extraction(phrec, pk_phhd, identity):
def information_extraction(phrec, pk_phhd):
"""
处理单张图片
:param phrec:图片信息
:param pk_phhd:案子主键
:param identity:处理批次标识
:return:记录类型，信息抽取结果
"""
img_path = image_util.get_img_path(phrec.cfjaddress)
if not img_path:
original_img_path = common_util.get_img_path(phrec.cfjaddress)
if not original_img_path:
img_url = ufile.get_private_url(phrec.cfjaddress)
if not img_url:
return None, None, None
img_path = image_util.save_to_local(img_url)
original_img_path = common_util.save_to_local(img_url)
img_path = common_util.get_processed_img_path(phrec.cfjaddress)
shutil.copy2(original_img_path, img_path)
# 尝试从二维码中获取高清图片
better_img_path, text = get_better_image_from_qrcode(img_path, phrec.cfjaddress)
@@ -129,17 +133,26 @@ def information_extraction(phrec, pk_phhd, identity):
)
ocr_text = None # 此处肯定不是出院记录，后续用不到
else:
target_image = model_util.det_book(img_path) # 识别文档区域并裁剪
dewarped_image = model_util.dewarp(target_image) # 去扭曲
angles = model_util.clas_orientation(dewarped_image)
rotated_img = image_util.rotate(dewarped_image, int(angles[0]))
split_results = image_util.split(rotated_img)
if image_util.is_photo(img_path):
book_img_path = model_util.det_book(img_path) # 识别文档区域并裁剪
dewarped_img_path = model_util.dewarp(book_img_path) # 去扭曲
else: # todo:也可能是图片，后续添加细分逻辑
dewarped_img_path = img_path
angles = model_util.clas_orientation(dewarped_img_path)
ocr_result = []
rotated_img = None
for angle in angles:
tmp_ocr_result = []
tmp_rotated_img = image_util.rotate(dewarped_img_path, int(angle))
split_results = image_util.split(tmp_rotated_img)
for split_result in split_results:
if split_result['img'] is None:
continue
a4_img = image_util.expand_to_a4_size(split_result['img'])
ocr_result += model_util.ocr(a4_img)
tmp_ocr_result += model_util.ocr(a4_img)
if len(tmp_ocr_result) > len(ocr_result):
ocr_result = tmp_ocr_result
rotated_img = tmp_rotated_img
ocr_text = common_util.ocr_result_to_text(ocr_result)
rec_type = model_util.clas_text(ocr_text) if ocr_text else None
if rec_type == '基本医保结算单':
@@ -158,7 +171,7 @@ def information_extraction(phrec, pk_phhd, identity):
now = common_util.get_default_datetime()
session = MysqlSession()
session.add(ZxIeResult(pk_phhd=pk_phhd, pk_phrec=phrec.pk_phrec, id=identity,
session.add(ZxIeResult(pk_phhd=pk_phhd, pk_phrec=phrec.pk_phrec, id=BATCH_ID,
cfjaddress=phrec.cfjaddress, content=result_json, create_time=now,
creator=HOSTNAME, update_time=now, updater=HOSTNAME))
session.commit()
@@ -175,8 +188,9 @@ def get_best_value_of_key(source, key):
values = source.get(key)
if values:
for value in values:
text = value.get("text")
probability = value.get("probability")
for v in value:
text = v.get("text")
probability = v.get("probability")
if text and probability > best_probability:
result = text
best_probability = probability
@@ -186,8 +200,9 @@ def get_best_value_of_key(source, key):
# 从keys中获取所有value组成list
def get_values_of_key(source, key):
result = []
value = source.get(key)
if value:
values = source.get(key)
if values:
for value in values:
for v in value:
v = v.get("text")
if v:
@@ -424,10 +439,12 @@ def photo_review(pk_phhd, name):
).all()
session.close()
# 同一批图的标识
identity = int(time.time())
modify_batch_id(uuid.uuid4().hex)
discharge_text = ''
for phrec in phrecs:
rec_type, ie_result, ocr_text = information_extraction(phrec, pk_phhd, identity)
processed_img_dir = common_util.get_processed_img_path('')
os.makedirs(processed_img_dir, exist_ok=True)
rec_type, ie_result, ocr_text = information_extraction(phrec, pk_phhd)
if rec_type == '基本医保结算单':
rec_result = settlement_result
elif rec_type == '出院记录':
@@ -439,10 +456,11 @@ def photo_review(pk_phhd, name):
rec_result = None
if rec_result is not None:
for key, value in ie_result.items():
if key == '页码':
rec_result[key].append(value) # 页码要区分来源，所以多包一层
else:
rec_result[key] += value
rec_result[key].append(value)
# 删除多余图片
if os.path.exists(processed_img_dir) and os.path.isdir(processed_img_dir):
shutil.rmtree(processed_img_dir)
settlement_data = settlement_task(pk_phhd, settlement_result)
discharge_data = discharge_task(pk_phhd, discharge_result)

52
util/common_util.py
View File

@@ -2,8 +2,12 @@ import logging
import os
from datetime import datetime
import requests
from opencc import OpenCC
from tenacity import retry, stop_after_attempt, wait_random
from log import PROJECT_ROOT
from photo_review import BATCH_ID
from util import string_util, model_util
@@ -255,3 +259,51 @@ def chinese_money_to_number(chinese_money_amount):
def traditional_to_simple_chinese(traditional_chinese):
converter = OpenCC('t2s')
return converter.convert(traditional_chinese)
def parse_img_url(url):
"""
解析图片url
:param url: 图片url
:return: 图片名称和图片后缀
"""
url = url.split('?')[0]
return os.path.basename(url)
@retry(stop=stop_after_attempt(3), wait=wait_random(1, 3), reraise=True,
after=lambda x: logging.warning('保存图片失败！'))
def save_to_local(img_url):
"""
保存图片到本地
:param img_url: 图片url
:return: 本地保存地址
"""
response = requests.get(img_url)
response.raise_for_status() # 检查响应状态码是否正常
save_path = get_tmp_img_path(parse_img_url(img_url))
with open(save_path, 'wb') as file:
file.write(response.content)
return save_path
def get_img_path(img_full_name):
save_path = get_tmp_img_path(img_full_name)
if os.path.exists(save_path):
return save_path
return None
def get_tmp_img_path(img_full_name):
return os.path.join(PROJECT_ROOT, 'tmp_img', img_full_name)
def get_processed_img_path(img_full_name):
return os.path.join(str(get_tmp_img_path(BATCH_ID)), img_full_name)
def parse_save_path(img_path):
img_full_name = os.path.basename(img_path)
img_name, img_ext = img_full_name.rsplit('.', 1)
return img_name, img_ext

75
util/image_util.py
View File

@@ -1,13 +1,10 @@
import logging
import math
import os
import cv2
import numpy
import requests
from tenacity import retry, stop_after_attempt, wait_random
from PIL import Image
from log import PROJECT_ROOT
from util import common_util
def capture(image, rectangle):
@@ -42,14 +39,14 @@ def split(img_path, ratio=1.414, overlap=0.05, x_compensation=3):
hw_ratio = height / width
wh_ratio = width / height
img_name, img_ext = parse_save_path(img_path)
img_name, img_ext = common_util.parse_save_path(img_path)
if hw_ratio > ratio: # 纵向过长
new_img_height = width * ratio
step = width * (ratio - overlap) # 偏移步长
for i in range(math.ceil(height / step)):
offset = round(step * i)
cropped_img = capture(image, [0, offset, width, offset + new_img_height])
split_path = get_save_path(f'{img_name}.split_{i}.{img_ext}')
split_path = common_util.get_processed_img_path(f'{img_name}.split_{i}.{img_ext}')
cv2.imwrite(split_path, cropped_img)
split_result.append({'img': split_path, 'x_offset': 0, 'y_offset': offset})
elif wh_ratio > ratio: # 横向过长
@@ -58,7 +55,7 @@ def split(img_path, ratio=1.414, overlap=0.05, x_compensation=3):
for i in range(math.ceil(width / step)):
offset = round(step * i)
cropped_img = capture(image, [offset, 0, offset + new_img_width, width])
split_path = get_save_path(f'{img_name}.split_{i}.{img_ext}')
split_path = common_util.get_processed_img_path(f'{img_name}.split_{i}.{img_ext}')
cv2.imwrite(split_path, cropped_img)
split_result.append({'img': split_path, 'x_offset': offset, 'y_offset': 0})
else:
@@ -111,8 +108,8 @@ def rotate(img_path, angle):
# 参数：原始图像 旋转参数 元素图像宽高
rotated = cv2.warpAffine(image, matrix, (new_width, new_height))
img_name, img_ext = parse_save_path(img_path)
rotated_path = get_save_path(f'{img_name}.rotate_{angle}.{img_ext}')
img_name, img_ext = common_util.parse_save_path(img_path)
rotated_path = common_util.get_processed_img_path(f'{img_name}.rotate_{angle}.{img_ext}')
cv2.imwrite(rotated_path, rotated)
return rotated_path
@@ -178,7 +175,7 @@ def expand_to_a4_size(img_path):
:return: 扩充后的图片NumPy数组和偏移量
"""
image = cv2.imread(img_path)
img_name, img_ext = parse_save_path(img_path)
img_name, img_ext = common_util.parse_save_path(img_path)
height, width = image.shape[:2]
x_offset, y_offset = 0, 0
hw_ratio = height / width
@@ -206,8 +203,9 @@ def expand_to_a4_size(img_path):
exp_img = numpy.zeros((y_offset, width, 3), dtype='uint8')
exp_img.fill(255)
image = numpy.vstack([exp_img, image, exp_img])
# todo:未拓展时不要生成新的图片
save_path = get_save_path(f'{img_name}.a4.{img_ext}')
else:
return img_path, 0, 0
save_path = common_util.get_processed_img_path(f'{img_name}.a4.{img_ext}')
cv2.imwrite(save_path, image)
return save_path, x_offset, y_offset
@@ -240,45 +238,16 @@ def combined(img1, img2):
return combined_img
def parse_img_url(url):
def is_photo(img_path):
"""
解析图片url
:param url: 图片url
:return: 图片名称和图片后缀
是否是拍照照片
:param img_path: 图片路径
:return: True:是照片；False:可能不是照片，也可能在传输过程中相机等信息丢失了
"""
url = url.split('?')[0]
return os.path.basename(url)
@retry(stop=stop_after_attempt(3), wait=wait_random(1, 3), reraise=True,
after=lambda x: logging.warning('保存图片失败！'))
def save_to_local(img_url):
"""
保存图片到本地
:param img_url: 图片url
:return: 本地保存地址
"""
response = requests.get(img_url)
response.raise_for_status() # 检查响应状态码是否正常
save_path = get_save_path(parse_img_url(img_url))
with open(save_path, 'wb') as file:
file.write(response.content)
return save_path
def get_img_path(img_full_name):
save_path = get_save_path(img_full_name)
if os.path.exists(save_path):
return save_path
return None
def get_save_path(img_full_name):
return os.path.join(PROJECT_ROOT, 'tmp_img', img_full_name)
def parse_save_path(img_path):
img_full_name = os.path.basename(img_path)
img_name, img_ext = img_full_name.rsplit('.', 1)
return img_name, img_ext
img = Image.open(img_path)
exif = img.getexif()
if exif:
# 271:相机制造商, 272:相机型号
if any(tag in exif for tag in (271, 272)):
return True
return False