Color transforms (brightness, contrast, hue) in Computer Vision - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

Output of brightness adjustment code

What will be the pixel value of the center pixel after applying this brightness increase code on a grayscale image with pixel value 100?

Computer Vision

import numpy as np
image = np.full((3,3), 100, dtype=np.uint8)
brightness_increase = 50
result = np.clip(image.astype(np.int16) + brightness_increase, 0, 255).astype(np.uint8)
center_pixel = result[1,1]
print(center_pixel)

A 150

B 100

C 50

D 255

💡 Hint

Adding brightness increases pixel values but caps at 255.

The original pixel value is 100. Adding 50 results in 150, which is within the 0-255 range, so the center pixel becomes 150.

🧠 Conceptual

intermediate

Effect of contrast adjustment on pixel values

If you multiply all pixel values in an image by 1.5 and then clip to 255, what happens to pixels originally at 200?

A They become 300

B They become 255 due to clipping

C They remain 200

D They become 133

💡 Hint

Multiplying 200 by 1.5 gives 300, but pixel values cannot exceed 255.

200 * 1.5 = 300, but pixel values are clipped to 255, so the final value is 255.

❓ Metrics

advanced

Measuring hue shift effect

Which metric best measures the difference in hue between two images after a hue shift?

A Cosine similarity on HSV hue channel

B Mean Squared Error on RGB channels

C Structural Similarity Index (SSIM)

D Pixel-wise absolute difference on grayscale

💡 Hint

Hue is a circular value best compared with angular similarity.

Cosine similarity on the hue channel in HSV color space captures angular differences in hue effectively.

🔧 Debug

advanced

Bug in contrast adjustment code

What error will this code raise? import numpy as np image = np.array([[50, 100], [150, 200]], dtype=np.uint8) contrast_factor = 1.2 result = image * contrast_factor print(result)

A OverflowError due to exceeding 255

B TypeError due to multiplying uint8 by float

C No error, but result is float array

D ValueError due to shape mismatch

💡 Hint

Multiplying a uint8 numpy array by a float returns a float array.

NumPy automatically upcasts the result to float64, so no error occurs but the result is float type.

❓ Model Choice

expert

Choosing model for color transform robustness

Which model architecture is best suited to learn robustness to brightness, contrast, and hue changes in images?

A Simple fully connected network without normalization

B Linear regression on raw pixel values

C Recurrent Neural Network processing pixel sequences

D Convolutional Neural Network with batch normalization and data augmentation

💡 Hint

Robustness to color changes benefits from convolution and normalization.

CNNs capture spatial features well, batch normalization stabilizes training, and data augmentation with color transforms improves robustness.