Prevent overflow when multiply numpy image array

Question

I am working with images in numpy and at some point I scale an image.

import scipy.misc        as msc
import numpy             as np
...
img_rgb = msc.imread(img_fn)
im_scaled = img_rgb * factor

The result sometimes looks ugly with bright regions showing black spots. This seems to be caused by numerical overflow of the 8bit image RGB pixel. Is there a way to apply a ceiling operator such that if the multiplication would be > 255 it is clipped to 255. (I am not interested in a floor function as I don't expect signal to become negative)

I know I can test every pixel in a loop, but the would not be following the numpy philosophy of array handling.

Any help is much appreciated.

Thanks, Gert

Peter · Accepted Answer · 2021-01-29 06:48:15Z

3

Use np.clip(x*float(factor), 0, 255).astype(np.uint8)

e.g.

x = np.array([120, 140], dtype=np.uint8)
factor = 2
result = np.clip(x*float(factor), 0, 255).astype(np.uint8)
> array([240, 255], dtype=uint8)

Note the float(factor) is important because if it is left as an int there will be overflow before the clip.

answered Jan 29, 2021 at 6:48

Peter

13.8k11 gold badges82 silver badges99 bronze badges

Sign up to request clarification or add additional context in comments.

Comments

heltonbiker · Accepted Answer · 2016-06-01 03:53:01Z

2

You can use numpy.clip(), it does exactely that.

However, the image is probably not going to be pretty either. Usually, what you want is to convert the image values from byte, in the range of [0,255], to float, in the range of [0,1] (even if implicitly), and apply gamma correction.

answered Jun 1, 2016 at 3:53

heltonbiker

27.7k30 gold badges151 silver badges270 bronze badges

3 Comments

Warren Weckesser Over a year ago

clip() doesn't really help with integer overflow. It is too late to clip after multiplying, because the overflow has already happened. You would have to upcast the input to, say, 16 bit integers or floating point, do the multiplication, and then clip.

Bi Rico Over a year ago

@WarrenWeckesser clip before the multiply, ie scaled = img.clip(0, 255 // factor) * factor.

heltonbiker Over a year ago

You commenters are right. That's why the implicit final advice, in the second paragraph, is to convert from byte to float, perform the operations, and only then convert back.

Collectives™ on Stack Overflow

Prevent overflow when multiply numpy image array

2 Answers 2

Comments

3 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

Comments

3 Comments

Your Answer

Sign up or log in

Post as a guest

Related