Save BGR image with cv2.imwrite

2
votes

I have a 2D numpy array which it's values are float between [-4, 3]. cv2.imshow shows this array as a BGR Image, but when I saved it with cv2.imwrite it was completely black. Then I found that I have to multiply the array to 255 to save it with imwrite, but in this case the image saved in RGB format, but I want to save the BGR image which shown by cv2.imshow. What should I do?

2
pythonnumpyopencv
Practically all image formats use RGB ordering. What format (file extension) do you want to use? – Mark Setchell
I'm saving the image as a PNG file – mhmmdrz92
PNG files are incapable of storing floats. You'll likely need TIFF or PFM. – Mark Setchell
Also, the ordering within any image file you save (PNG, BMP, GIF, JPG) is not really relevant because OpenCV converts it on-the-fly as it reads or writes. The order is standardised to RGB so other applications can correctly interpret your images. – Mark Setchell
Images in floating point format must have values in range [0,1]. Ones with 8 bit unsigned need to be in range [0,255]. Multiplying values in range [-4,3] by 255 will still leave you with invalid values (or worse, cause overflow, completely trashing the data.) – Dan Mašek

2 Answers

1
votes

So, basically you want to convert your image from RGB to BGR image.

This can be done by using cv2.cvtColor() function.

result_BGR = cv2.cvtColor(RGB_image, cv2.COLOR_RGB2BGR)
cv2.imwrite('PATH', result_BGR)
0
votes

First of all you have to adjust the value of all arrays. The pixels are between -4 and 3 so you have to do this:

img = img - min_val
img = img*255.0/(max_val - min_val)

which in your case it would be like:

img = img+4
img = img*255/7.0

then convert your img to 8bit unsinged int and save it with imwrite (no need to mess with BGR or RGB, opencv handles it by itself)