Automated Inspection with Computer Vision/

...

/

Smoothing

Smoothing

Learn about the types of image smoothing, and when to use them.

We'll cover the following...

Smoothing an image

1.

When do we need to smooth an image?

Show Answer
Q1 / Q1
Did you find this helpful?

What do we mean by noise?

Noise in an image is a random quantity that adds to the signal, corrupting the pixel’s gray level or color. As a function of the imaging system that generates our images and the lighting conditions, noise could be so severe as to prevent any automated inspection task.

Let’s see this in action. Consider the image below, severely affected by uncorrelated Gaussian noise.

Press + to interact
An image severely affected by uncorrelated Gaussian noise
An image severely affected by uncorrelated Gaussian noise

Let’s convert this image to grayscale, apply various uniform thresholds, and see the resulting masks.

Press + to interact
import cv2
original_img = cv2.imread('./images/fruits/bananas_black1b_noise.jpg')
grayscale_img = cv2.cvtColor(original_img, cv2.COLOR_BGR2GRAY)
# Apply various uniform thresholds
thresholds = [60, 90, 105, 120, 150]
for threshold_ndx in range(len(thresholds)):
    threshold = thresholds[threshold_ndx]
    retval, mask = cv2.threshold(grayscale_img, threshold, 255, cv2.THRESH_BINARY)
    # Annotate the mask with the applied threshold value
    cv2.putText(mask, str(threshold), (0, 60), cv2.FONT_HERSHEY_SIMPLEX, 
    1.0, 255, 2)
    cv2.imwrite(f'./output/{threshold_ndx + 1}_mask.png', mask)
cv2.imwrite('./output/0_original.png', original_img)

In line 10, we call the cv2.threshold() function, defined at line 7, with various threshold values.

We can see that we can’t highlight the whole surface of the bananas without having some spurious pixels ...