Image thresholding (binary, adaptive, Otsu) in Computer Vision - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

Output of binary thresholding with OpenCV

What is the shape and unique values of the output image after applying binary thresholding with threshold=127 on a grayscale image of shape (100, 100) with pixel values ranging from 0 to 255?

Computer Vision

import cv2
import numpy as np
img = np.random.randint(0, 256, (100, 100), dtype=np.uint8)
_, thresh_img = cv2.threshold(img, 127, 255, cv2.THRESH_BINARY)
unique_vals = np.unique(thresh_img)
output_shape = thresh_img.shape
print(output_shape, unique_vals)

A (100, 100, 3) and [0 255]

B (100, 100) and [0 255]

C (100, 100) and [0 127]

D (100, 100) and [127 255]

💡 Hint

Binary thresholding converts pixels to either 0 or max value based on threshold.

Binary thresholding sets pixels above threshold to max value (255) and below to 0, keeping the image shape unchanged.

🧠 Conceptual

intermediate

Understanding adaptive thresholding

Which statement best describes adaptive thresholding compared to global binary thresholding?

A Adaptive thresholding requires the image to be color, not grayscale.

B Adaptive thresholding uses a fixed threshold for the entire image.

C Adaptive thresholding calculates threshold for small regions, handling varying lighting.

D Adaptive thresholding always produces a binary image with only zeros.

💡 Hint

Think about how lighting changes across an image.

Adaptive thresholding computes thresholds locally for small areas, making it good for images with uneven lighting.

❓ Metrics

advanced

Evaluating Otsu's thresholding output

After applying Otsu's thresholding on a bimodal grayscale image, what metric best indicates the quality of the threshold?

A Maximizing inter-class variance between foreground and background

B Minimizing mean squared error of pixel intensities

C Maximizing the number of pixels set to zero

D Minimizing the image histogram entropy

💡 Hint

Otsu's method tries to separate two classes clearly.

Otsu's method finds the threshold that maximizes the variance between two classes (foreground and background), improving separation.

🔧 Debug

advanced

Identifying error in adaptive thresholding code

What error will this code raise? import cv2 import numpy as np img = np.random.randint(0, 256, (50, 50), dtype=np.uint8) thresh = cv2.adaptiveThreshold(img, 255, cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY, 2, 5)

A cv2.error: blockSize must be odd and greater than 1

B TypeError: unsupported operand type(s) for +: 'int' and 'str'

C No error, code runs successfully

D ValueError: image must be 3-channel color

💡 Hint

Check the blockSize parameter requirements in adaptiveThreshold.

The blockSize parameter must be an odd number greater than 1; 2 is invalid and causes an OpenCV error.

❓ Model Choice

expert

Choosing thresholding method for uneven lighting

You have a grayscale image with strong shadows and bright spots. Which thresholding method is best to segment the foreground accurately?

A Global binary thresholding with a fixed threshold

B Otsu's thresholding

C No thresholding, use raw pixel values

D Adaptive thresholding with Gaussian weighting

💡 Hint

Consider how local lighting variations affect thresholding.

Adaptive thresholding with Gaussian weighting adjusts thresholds locally, handling uneven lighting better than global or Otsu's methods.