统一照片脱敏的图片流转方式

photo_mask/__init__.py

@@ -1,5 +1,3 @@
-from paddleocr import PaddleOCR
-
 """
 项目配置
 """
@@ -40,13 +38,3 @@ SIMILAR_CHAR = {
     "侯": ["候"],
     "宇": ["字"],
 }
-
-# 如果不希望识别出空格，可以设置use_space_char=False。做此项设置一定要测试，2.7.3版本此项设置有bug，会导致识别失败
-OCR = PaddleOCR(
-    gpu_id=0,
-    show_log=False,
-    det_db_thresh=0.1,
-    det_db_box_thresh=0.3,
-    det_limit_side_len=1248,
-    drop_score=0.3
-)

photo_mask/auto_photo_mask.py

@@ -1,6 +1,5 @@
 import logging.config
 import re
-import tempfile
 import time
 from time import sleep
 
@@ -10,9 +9,9 @@ from sqlalchemy import update, and_
 from db import MysqlSession
 from db.mysql import ZxPhrec, ZxPhhd
 from log import HOSTNAME
-from photo_mask import OCR, PHHD_BATCH_SIZE, SLEEP_MINUTES, NAME_KEYS, ID_CARD_NUM_KEYS, SIMILAR_CHAR
+from photo_mask import PHHD_BATCH_SIZE, SLEEP_MINUTES, NAME_KEYS, ID_CARD_NUM_KEYS, SIMILAR_CHAR
 from ucloud import BUCKET, ufile
-from util import image_util, common_util
+from util import image_util, common_util, model_util
 
 
 def find_boxes(content, layout, offset=0, length=None, improve=False, image_path=None, extra_content=None):
@@ -55,14 +54,15 @@ 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)
         # 截图时偏大一点
         capture_box = common_util.zoom_rectangle(box, 0.2)
         captured_image = image_util.capture(image, capture_box)
-        with tempfile.NamedTemporaryFile(suffix=".jpg", delete=False) as temp_file:
-            captured_image, offset_x, offset_y = image_util.expand_to_a4_size(captured_image)
-            cv2.imwrite(temp_file.name, captured_image)
+        captured_image_path = image_util.get_save_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:
-            layouts = common_util.get_ocr_layout(OCR, temp_file.name)
+            layouts = common_util.ocr_result_to_layout(model_util.ocr(captured_a4_img_path))
         except TypeError:
             # 如果是类型错误，大概率是没识别到文字
             layouts = []
@@ -86,22 +86,17 @@ def find_boxes(content, layout, offset=0, length=None, improve=False, image_path
                             temp_box[3] + capture_box[1] - offset_y,
                         ])
                         break
-        common_util.delete_temp_file(temp_file.name)
 
     if not boxes:
         boxes.append(box)
     return boxes
 
 
-def get_mask_layout(image, name, id_card_num):
-    with tempfile.NamedTemporaryFile(delete=False, suffix=".jpg") as temp_file:
-        cv2.imwrite(temp_file.name, image)
-
+def get_mask_layout(img_path, name, id_card_num):
     result = []
     try:
         try:
-            layouts = common_util.get_ocr_layout(OCR, temp_file.name)
-            # layouts = OCR.parse({"doc": temp_file.name})["layout"]
+            layouts = common_util.ocr_result_to_layout(model_util.ocr(img_path))
         except TypeError:
             # 如果是类型错误，大概率是没识别到文字
             layouts = []
@@ -135,12 +130,12 @@ def get_mask_layout(image, name, id_card_num):
                 find_id_card_num_by_key = True
                 matches = re.findall(r, layout[1])
                 for match in matches:
-                    result += find_boxes(match, layout, improve=True, image_path=temp_file.name, extra_content=r)
+                    result += find_boxes(match, layout, improve=True, image_path=img_path, extra_content=r)
                     find_name_by_key = False
                     break
 
                 if id_card_num in layout[1]:
-                    result += find_boxes(id_card_num, layout, improve=True, image_path=temp_file.name)
+                    result += find_boxes(id_card_num, layout, improve=True, image_path=img_path)
                     find_id_card_num_by_key = False
 
                 def _find_boxes_by_keys(keys):
@@ -163,8 +158,6 @@ def get_mask_layout(image, name, id_card_num):
     except Exception as e:
         logging.error("涂抹时出错！", exc_info=e)
         return result
-    finally:
-        common_util.delete_temp_file(temp_file.name)
 
 
 def handle_image_for_mask(split_result):
@@ -174,10 +167,12 @@ def handle_image_for_mask(split_result):
     return expand_img, split_result["x_offset"], split_result["y_offset"]
 
 
-def mask_photo(img_url, name, id_card_num, color=(255, 255, 255)):
-    def _mask(i, n, icn, c):
+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)
         do_mask = False
-        split_results = image_util.split(i)
+        split_results = image_util.split(ip)
         for split_result in split_results:
             to_mask_img, x_offset, y_offset = handle_image_for_mask(split_result)
             results = get_mask_layout(to_mask_img, n, icn)
@@ -193,27 +188,27 @@ def mask_photo(img_url, 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)
-        return do_mask, i
+        masked_path = image_util.get_save_path(f'{img_name}.mask.{img_ext}')
+        cv2.imwrite(masked_path, i)
+        return do_mask, masked_path
 
-    # 打开图片
-    image = image_util.read(img_url)
-    original_image = image
-    is_masked, image = _mask(image, name, id_card_num, color)
+    original_image = img_path
+    is_masked, img_path = _mask(img_path, name, id_card_num, color)
     if not is_masked:
         # 如果没有涂抹，可能是图片方向不对
-        angles = image_util.parse_rotation_angles(image)
+        angles = model_util.clas_orientation(img_path)
         angle = angles[0]
         if angle != "0":
-            image = image_util.rotate(image, int(angle))
-            is_masked, image = _mask(image, name, id_card_num, color)
+            img_path = image_util.rotate(img_path, int(angle))
+            is_masked, img_path = _mask(img_path, name, id_card_num, color)
         if not is_masked:
             # 如果旋转后也没有涂抹，恢复原来的方向
-            image = original_image
+            img_path = original_image
         else:
             # 如果旋转有效果，打一个日志
             logging.info(f"图片旋转了{angle}°")
 
-    return is_masked, image
+    return is_masked, img_path
 
 
 def photo_mask(pk_phhd, name, id_card_num):
@@ -227,17 +222,14 @@ def photo_mask(pk_phhd, name, id_card_num):
         img_url = ufile.get_private_url(phrec.cfjaddress)
         if not img_url:
             continue
-
-        is_masked, image = mask_photo(img_url, name, id_card_num)
+        img_path = image_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
         if is_masked:
             ufile.copy_file(BUCKET, phrec.cfjaddress, "drg2015", phrec.cfjaddress)
-
-            with tempfile.NamedTemporaryFile(delete=False, suffix=".jpg") as temp_file:
-                cv2.imwrite(temp_file.name, image)
             try:
-                ufile.upload_file(phrec.cfjaddress, temp_file.name)
+                ufile.upload_file(phrec.cfjaddress, image)
                 session = MysqlSession()
                 update_flag = (update(ZxPhrec).where(ZxPhrec.pk_phrec == phrec.pk_phrec).values(
                     paint_user=HOSTNAME,
@@ -248,7 +240,7 @@ def photo_mask(pk_phhd, name, id_card_num):
             except Exception as e:
                 logging.error("上传图片出错", exc_info=e)
             finally:
-                common_util.delete_temp_file(temp_file.name)
+                common_util.delete_temp_file(image)
 
 
 def main():

photo_review/auto_photo_review.py

@@ -111,6 +111,8 @@ def information_extraction(phrec, pk_phhd, identity):
     img_path = image_util.get_img_path(phrec.cfjaddress)
     if not 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)
 
     # 尝试从二维码中获取高清图片

services/paddle_services/ocr.py

@@ -7,7 +7,9 @@ from log import LOGGING_CONFIG
 from utils import process_request
 
 app = Flask(__name__)
-OCR = PaddleOCR(use_angle_cls=False, show_log=False, gpu_id=0, det_db_box_thresh=0.3)
+# 如果不希望识别出空格，可以设置use_space_char=False。做此项设置一定要测试，2.7.3版本此项设置有bug，会导致识别失败
+OCR = PaddleOCR(use_angle_cls=False, show_log=False, det_db_thresh=0.1, det_db_box_thresh=0.3, det_limit_side_len=1248,
+                drop_score=0.3)
 
 
 @app.route('/', methods=['POST'])