diff --git a/paddleocr.py b/paddleocr.py
index dc92cbf6b7..36980aec44 100644
--- a/paddleocr.py
+++ b/paddleocr.py
@@ -516,7 +516,19 @@ def img_decode(content: bytes):
     return cv2.imdecode(np_arr, cv2.IMREAD_UNCHANGED)
 
 
-def check_img(img):
+def check_img(img, alpha_color=(255, 255, 255)):
+    """
+    Check the image data. If it is another type of image file, try to decode it into a numpy array.
+    The inference network requires three-channel images, So the following channel conversions are done
+        single channel image: Gray to RGB R←Y,G←Y,B←Y
+        four channel image: alpha_to_color
+    args:
+        img: image data
+            file format: jpg, png and other image formats that opencv can decode, as well as gif and pdf formats
+            storage type: binary image, net image file, local image file
+        alpha_color: Background color in images in RGBA format
+        return: numpy.array (h, w, 3)
+    """
     if isinstance(img, bytes):
         img = img_decode(img)
     if isinstance(img, str):
@@ -550,9 +562,12 @@ def check_img(img):
         if img is None:
             logger.error("error in loading image:{}".format(image_file))
             return None
+    # single channel image array.shape:h,w
     if isinstance(img, np.ndarray) and len(img.shape) == 2:
         img = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)
-
+    # four channel image array.shape:h,w,c
+    if isinstance(img, np.ndarray) and len(img.shape) == 3 and img.shape[2] == 4:
+        img = alpha_to_color(img, alpha_color)
     return img
 
 
@@ -638,7 +653,7 @@ def ocr(self, img, det=True, rec=True, cls=True, bin=False, inv=False, alpha_col
                 'Since the angle classifier is not initialized, it will not be used during the forward process'
             )
 
-        img = check_img(img)
+        img = check_img(img, alpha_color)
         # for infer pdf file
         if isinstance(img, list):
             if self.page_num > len(img) or self.page_num == 0:
@@ -648,7 +663,6 @@ def ocr(self, img, det=True, rec=True, cls=True, bin=False, inv=False, alpha_col
             imgs = [img]
 
         def preprocess_image(_image):
-            _image = alpha_to_color(_image, alpha_color)
             if inv:
                 _image = cv2.bitwise_not(_image)
             if bin:
@@ -755,8 +769,8 @@ def __init__(self, **kwargs):
         logger.debug(params)
         super().__init__(params)
 
-    def __call__(self, img, return_ocr_result_in_table=False, img_idx=0):
-        img = check_img(img)
+    def __call__(self, img, return_ocr_result_in_table=False, img_idx=0, alpha_color=(255, 255, 255)):
+        img = check_img(img, alpha_color)
         res, _ = super().__call__(
             img, return_ocr_result_in_table, img_idx=img_idx)
         return res