从 BGR 到灰度的 OpenCV 颜色转换错误

Question

I am trying to convert an image from BGR to grayscale format using this code:

img = cv2.imread('path//to//image//file')
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

This seems to be working fine: I checked the data type of the img variable which turns out to be numpy ndarray and shape to be (100,80,3) . However if I give an image of a native numpy ndarray data type with same dimensions of the input of the cvtColor function, it gives me the following error:

Error: Assertion failed (depth == 0 || depth == 2 || depth == 5) in cv::cvtColor, file D:\Build\OpenCV\opencv-3.4.1\modules\imgproc\src\color.cpp, line 11109
cv2.error: OpenCV(3.4.1) D:\Build\OpenCV\opencv-3.4.1\modules\imgproc\src\color.cpp:11109: error: (-215) depth == 0 || depth == 2 || depth == 5 in function cv::cvtColor

The code for the second case is (making a custom np.ndarray over here):

img = np.full((100,80,3), 12)
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) 

Can anyone clarify what is the reason for this error and how to rectify it?

Answer 1

This is because your numpy array is not made up of the right data type. By default makes an array of type np.int64 (64 bit), however, cv2.cvtColor() requires 8 bit ( np.uint8 ) or 16 bit ( np.uint16 ). To correct this change your np.full() function to include the data type:

img = np.full((100,80,3), 12, np.uint8)

Answer 2

The error occured because the datatype of numpy array returned by cv2.imread is uint8 , which is different from the datatype of numpy array returned by np.full() . To make the data-type as uint8, add the dtype parameter-

img = np.full((100,80,3), 12, dtype = np.uint8)
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)

Answer 3

It may be easier to initialize new numpy array with initial image as source and dtype=np.uint8 :

import numpy as np    
img = cv2.imread('path//to//image//file')    
img = np.array(img, dtype=np.uint8)
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

Answer 4

imagine you have a function called preprocessing() that preprocess your images with cv2, if you try to apply it as:

data = np.array(list(map(preprocessing,data)))

it won't work and that because np.array creates int64and you are trying to assign np.uint8 to it, what you should do instead is adding dtype parameter as follow:

data = np.array(list(map(preprocessing,data)), dtype = np.uint8)