因pillow在linux中有问题难以解决，更换图片处理库为openCV

photo_review/photo_review.py

@@ -5,14 +5,14 @@ import os
 import sys
 import tempfile
 import time
-from io import BytesIO
+import urllib.request
 
+import cv2
+import numpy as np
 import paddle
-import requests
 
 sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
 
-from PIL import Image
 from time import sleep
 from sqlalchemy import update
 from config.keys import PATIENT_NAME, ADMISSION_DATE, DISCHARGE_DATE, MEDICAL_EXPENSES, PERSONAL_CASH_PAYMENT, \
@@ -35,16 +35,18 @@ from ucloud import ucloud
 
 
 # 获取图片
-def open_image_from_url(url):
+def open_image(img_path):
-    if url.startswith("http"):
+    if img_path.startswith("http"):
-        # 发送HTTP请求获取图片数据
+        # 发送HTTP请求并获取图像数据
-        response = requests.get(url)
+        resp = urllib.request.urlopen(img_path)
-        # 将响应内容转化为BytesIO对象，以便PIL处理
+        # 将数据读取为字节流
-        image_stream = BytesIO(response.content)
+        image_data = resp.read()
-        # 使用PIL的Image.open方法打开图像
+        # 将字节流转换为NumPy数组
-        image = Image.open(image_stream)
+        image_np = np.frombuffer(image_data, np.uint8)
+        # 解码NumPy数组为OpenCV图像格式
+        image = cv2.imdecode(image_np, cv2.IMREAD_COLOR)
     else:
-        image = Image.open(url)
+        image = cv2.imread(img_path)
     return image
 
 
@@ -52,31 +54,37 @@ def open_image_from_url(url):
 def split_image(img_path, max_ratio=2.82, best_ration=1.41, overlap=0.05):
     split_result = []
     # 打开图片
-    img = open_image_from_url(img_path)
+    img = open_image(img_path)
     # 获取图片的宽度和高度
-    width, height = img.size
+    height, width = img.shape[:2]
     # 计算宽高比
     ratio = max(width, height) / min(width, height)
     # 检查是否需要裁剪
     if ratio > max_ratio:
         # 确定裁剪的尺寸，保持长宽比，以较短边为基准
         new_ratio = best_ration - overlap
-        if width < height:  # 高度是较长边
+        if width < height:
+            # 高度是较长边
+            cropped_width = width * best_ration
             for i in range(math.ceil(height / (width * new_ratio))):
                 offset = round(width * new_ratio * i)
-                cropped_img = img.crop((0, offset, width, round(offset + width * best_ration)))
+                # 参数形式为[y1:y2, x1:x2]
-                # 统一转为RGB，这样可以正确保存为jpg格式
+                cropped_img = img[offset:round(offset + cropped_width), 0:width]
-                cropped_img = cropped_img.convert("RGB")
                 split_result.append({"img": cropped_img, "x_offset": 0, "y_offset": offset})
-        else:  # 宽度是较长边
+            # 最后一次裁剪时不足的部分填充黑色
+            last_img = split_result[-1]["img"]
+            split_result[-1]["img"] = cv2.copyMakeBorder(last_img, 0, round(cropped_width - last_img.shape[0]), 0, 0, cv2.BORDER_CONSTANT, value=(0, 0, 0))
+        else:
+            # 宽度是较长边
+            cropped_height = height * best_ration
             for i in range(math.ceil(width / (height * new_ratio))):
                 offset = round(height * new_ratio * i)
-                cropped_img = img.crop((offset, 0, round(offset + height * best_ration), height))
+                cropped_img = img[0:height, offset:round(offset + cropped_height)]
-                # 统一转为RGB，这样可以正确保存为jpg格式
-                cropped_img = cropped_img.convert("RGB")
                 split_result.append({"img": cropped_img, "x_offset": offset, "y_offset": 0})
+            # 最后一次裁剪时不足的部分填充黑色
+            last_img = split_result[-1]["img"]
+            split_result[-1]["img"] = cv2.copyMakeBorder(last_img, 0, 0, 0, round(cropped_height - last_img.shape[1]), cv2.BORDER_CONSTANT, value=(0, 0, 0))
     else:
-        img = img.convert("RGB")
         split_result.append({"img": img, "x_offset": 0, "y_offset": 0})
     return split_result
 
@@ -99,8 +107,7 @@ def information_extraction(ie, phrecs):
             split_result = split_image(pic_path)
             for img in split_result:
                 with tempfile.NamedTemporaryFile(delete=False, suffix=".jpg") as temp_file:
-                    img["img"].save(temp_file.name)
+                    cv2.imwrite(temp_file.name, img["img"])
-
                     docs.append({"doc": temp_file.name})
                     doc_phrecs.append({"phrec": phrec, "x_offset": img["x_offset"], "y_offset": img["y_offset"]})
     if not docs:

test.py

@@ -1,5 +0,0 @@
-from photo_review.photo_review import open_image_from_url
-
-if __name__ == '__main__':
-    image = open_image_from_url("./PH20240529000194_1_075936_1.png")
-    image.save("test.png")

visual_model_test/visual_model_test.py

@@ -6,6 +6,8 @@ import tempfile
 import time
 from pprint import pprint
 
+import cv2
+
 from photo_review.photo_review import split_image
 
 sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
@@ -48,7 +50,7 @@ def visual_model_test(model_type, test_img, task_path, schema):
         doc_parser = DocParser(layout_analysis=False)
         for img in imgs:
             with tempfile.NamedTemporaryFile(delete=False, suffix=".jpg") as temp_file:
-                img["img"].save(temp_file.name)
+                cv2.imwrite(temp_file.name, img["img"])
                 temp_files_paths.append(temp_file.name)
                 parsed_doc = doc_parser.parse({"doc": temp_file.name}, expand_to_a4_size=True)
                 if img["x_offset"] or img["y_offset"]:
@@ -67,7 +69,7 @@ def visual_model_test(model_type, test_img, task_path, schema):
         temp_files_paths = []
         for img in split_result:
             with tempfile.NamedTemporaryFile(delete=False, suffix=".jpg") as temp_file:
-                img["img"].save(temp_file.name)
+                cv2.imwrite(temp_file.name, img["img"])
                 temp_files_paths.append(temp_file.name)
                 docs.append({"doc": temp_file.name})