增加长图切割

format_convert/convert_tree.py

@@ -118,18 +118,71 @@ class _Image:
     def get_text(self):
         return
 
+    def imageSlice(self,image_np):
+        '''
+        slice the image if the height is to large
+        :return:
+        '''
+        if image_np is None:
+            return []
+
+        # 整体分辨率限制
+        if image_np.shape[0] > 3000 and image_np.shape[1] < 2000:
+            _sum = np.average(image_np,axis=1)
+
+            list_white_line = []
+            list_ave = list(_sum)
+            for _i in range(len(list_ave)):
+                if (list_ave[_i]>250).all():
+                    list_white_line.append(_i)
+            set_white_line = set(list_white_line)
+            width = image_np.shape[1]
+            height = image_np.shape[0]
+            list_images = []
+            _begin = 0
+            _end = 0
+            while 1:
+                if _end>height:
+                    break
+                _end+= width
+                while 1:
+                    if _begin in set_white_line:
+                        break
+                    if _begin>height:
+                        break
+                    _begin += 1
+                _image = image_np[_begin:_end,...]
+                list_images.append(_image)
+                _begin = _end
+            print("image slice into %d parts"%(len(list_images)))
+            return list_images
+        return [image_np]
+
+
+
     def convert(self):
         # 二进制转numpy
         # image_np = Image.open(io.BytesIO(self.content))
         # image_np = cv2.cvtColor(np.asarray(image_np), cv2.COLOR_RGB2BGR)
         image_np = cv2.imread(self.path)
 
-        obj_list = image_process(image_np, self.path, self.is_from_pdf, self.is_from_docx, use_ocr=True)
-        if judge_error_code(obj_list):
-            self.error_code = obj_list
-            return
-        for obj in obj_list:
-            self.add_child(obj)
+        list_images = self.imageSlice(image_np)
+        # print(len(list_images))
+        # return
+
+
+        _add_y = 0
+        for _image in list_images:
+            obj_list = image_process(_image, self.path, self.is_from_pdf, self.is_from_docx, use_ocr=True)
+            if judge_error_code(obj_list):
+                self.error_code = obj_list
+            else:
+                list_y = []
+                for obj in obj_list:
+                    obj.y += _add_y
+                    list_y.append(obj.y)
+                    self.add_child(obj)
+                _add_y = max(list_y)
 
 
 class _Table:

isr/isr_interface.py

@@ -100,7 +100,7 @@ def detect_seal(image_np, model):
     # inference data
     with sess1.as_default():
         with sess1.graph.as_default():
-            out_boxes, out_scores, out_classes = model.predict([image_resize, image_shape], batch_size=1000, steps=1)
+            out_boxes, out_scores, out_classes = model.predict([image_resize, image_shape], steps=2)
     if int(out_boxes.shape[0]) == 0:
         log("there is no seal!")
         return image_np, [], []