Color transforms (brightness, contrast, hue) in Computer Vision - Model Pipeline Trace

Model Pipeline - Color transforms (brightness, contrast, hue)

This pipeline shows how an image's colors are changed step-by-step by adjusting brightness, contrast, and hue. These changes help a model learn better by seeing different versions of the same image.

Data Flow - 4 Stages

1Input Image

1 image x 224 height x 224 width x 3 color channels→Original image loaded with RGB colors→1 image x 224 x 224 x 3

A photo of a red apple with green leaves

↓

2Brightness Adjustment

1 image x 224 x 224 x 3→Add a small value to each pixel to make image lighter or darker→1 image x 224 x 224 x 3

Apple image becomes slightly brighter, colors look lighter

↓

3Contrast Adjustment

1 image x 224 x 224 x 3→Stretch or shrink pixel values around middle gray to increase or decrease contrast→1 image x 224 x 224 x 3

Apple image colors become more vivid or more faded

↓

4Hue Adjustment

1 image x 224 x 224 x 3→Shift colors around the color wheel to change overall tint→1 image x 224 x 224 x 3

Apple image colors shift slightly from red to orange or purple

Training Trace - Epoch by Epoch

Loss
1.0 |***************
0.8 |************   
0.6 |********      
0.4 |******        
0.2 |***           
0.0 +-------------
      1 2 3 4 5 Epochs

Epoch	Loss ↓	Accuracy ↑	Observation
1	0.85	0.55	Model starts learning with high loss and moderate accuracy
2	0.65	0.70	Loss decreases as model adjusts to color variations
3	0.50	0.78	Model improves recognizing images despite color changes
4	0.40	0.85	Better generalization with color transform augmentations
5	0.35	0.88	Training converges with stable loss and high accuracy

Prediction Trace - 5 Layers

Layer 1: Input Image

Layer 2: Brightness Adjustment

Layer 3: Contrast Adjustment

Layer 4: Hue Adjustment

Layer 5: Final Output Image

Model Quiz - 3 Questions

What happens to pixel values during brightness adjustment?

A A fixed value is added to all pixels

B Pixels are multiplied by a factor

C Colors are shifted around the color wheel

D Pixels are replaced with random noise

Key Insight

Applying brightness, contrast, and hue changes creates new image versions that help the model learn to recognize objects under different lighting and color conditions. This improves the model's ability to generalize to real-world images.

Practice

1. What does increasing the brightness of an image do to its pixel values?

easy

A. It makes pixel values larger, making the image lighter

B. It rotates the colors around the color wheel

C. It stretches the difference between pixel values

D. It converts the image to grayscale

Solution

1. Step 1: Understand brightness effect on pixels

   Brightness changes add or multiply pixel values to make them lighter or darker.

2. Step 2: Identify what increasing brightness means

   Increasing brightness means making pixel values larger, so the image looks lighter.

3. Final Answer:

   It makes pixel values larger, making the image lighter -> Option A

4. Quick Check:

   Brightness increase = lighter pixels [OK]

Hint: Brightness changes scale pixel values up or down [OK]

Common Mistakes:

• Confusing brightness with contrast
• Thinking brightness rotates colors
• Assuming brightness converts to grayscale

2. Which of the following is the correct way to increase contrast of an image using a simple formula?

easy

A. new_pixel = (old_pixel - 128) * 1.5 + 128

B. new_pixel = old_pixel + 50

C. new_pixel = old_pixel / 2

D. new_pixel = old_pixel * 0

Solution

1. Step 1: Recall contrast adjustment formula

   Contrast changes stretch pixel values around the midpoint (usually 128 for 8-bit images).

2. Step 2: Identify correct formula

   The formula (old_pixel - 128) * factor + 128 increases contrast by scaling differences from midpoint.

3. Final Answer:

   new_pixel = (old_pixel - 128) * 1.5 + 128 -> Option A

4. Quick Check:

   Contrast = scale around midpoint [OK]

Hint: Contrast scales difference from midpoint, not just add values [OK]

Common Mistakes:

• Adding a constant increases brightness, not contrast
• Dividing pixel values reduces brightness
• Multiplying by zero makes image black

3. Given this Python code snippet using OpenCV, what will be the effect on the image stored in img?

import cv2
import numpy as np
img = cv2.imread('image.jpg')
hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
hsv[:, :, 0] = (hsv[:, :, 0] + 30) % 180
img_shifted = cv2.cvtColor(hsv, cv2.COLOR_HSV2BGR)

medium

A. The image is converted to grayscale

B. The image contrast is increased by 30%

C. The image brightness is increased by 30 units

D. The image hue is shifted by 30 degrees on the color wheel

Solution

1. Step 1: Understand HSV color space and hue channel

   HSV separates color into Hue, Saturation, and Value. Hue channel controls color angle on the wheel.

2. Step 2: Analyze code effect on hue channel

   The code adds 30 to the hue channel and wraps around with modulo 180, shifting colors on the wheel.

3. Final Answer:

   The image hue is shifted by 30 degrees on the color wheel -> Option D

4. Quick Check:

   Hue shift = color rotation [OK]

Hint: Hue changes rotate colors in HSV space [OK]

Common Mistakes:

• Confusing hue shift with brightness or contrast
• Ignoring modulo wrap-around for hue
• Assuming color space conversion changes brightness

4. The following code attempts to increase image contrast but does not work as expected. What is the main error?

def increase_contrast(img):
    mean = img.mean()
    return (img - mean) * 1.2 + mean

result = increase_contrast(image)

medium

A. The image should be converted to grayscale first

B. The multiplication factor should be less than 1

C. The mean should be computed per channel, not overall

D. The code should add 1.2 instead of multiplying

Solution

1. Step 1: Understand contrast adjustment per channel

   Contrast should be adjusted per color channel to preserve color balance.

2. Step 2: Identify error in mean calculation

   The code calculates mean over entire image, mixing channels, causing wrong contrast scaling.

3. Final Answer:

   The mean should be computed per channel, not overall -> Option C

4. Quick Check:

   Contrast per channel = correct scaling [OK]

Hint: Calculate mean per channel for contrast adjustment [OK]

Common Mistakes:

• Using single mean for all channels
• Using factor less than 1 reduces contrast
• Adding constant changes brightness, not contrast

5. You want to apply a color transform pipeline to an image that first increases brightness by 20%, then shifts hue by 45 degrees, and finally increases contrast by 30%. Which order of operations is best to preserve natural colors?

hard

A. Increase contrast -> Shift hue -> Increase brightness

B. Increase brightness -> Increase contrast -> Shift hue

C. Shift hue -> Increase brightness -> Increase contrast

D. Shift hue -> Increase contrast -> Increase brightness

Solution

1. Step 1: Consider effect of brightness before contrast

   Increasing brightness first ensures pixel values are lifted before stretching differences with contrast.

2. Step 2: Apply hue shift last to avoid color distortion

   Hue shift should be last to rotate colors after brightness and contrast adjustments.

3. Step 3: Confirm order preserves natural colors

   Brightness -> Contrast -> Hue maintains natural look by adjusting lightness and contrast before color rotation.

4. Final Answer:

   Increase brightness -> Increase contrast -> Shift hue -> Option B

5. Quick Check:

   Brightness then contrast then hue = natural colors [OK]

Hint: Adjust brightness and contrast before hue shift [OK]

Common Mistakes:

• Shifting hue before brightness causes unnatural colors
• Increasing contrast before brightness can clip pixels
• Applying hue shift first distorts brightness and contrast