Image Transformations and Filtering

# CHAPTER 5

Image Transformations and Filtering #

1. Introduction #

2. Learning Objectives #

• Understand affine transformations (Scaling, Rotation, Translation).

• Explain what a Kernel/Convolution matrix is.

• Understand how Blurring (Smoothing) reduces high-frequency noise.

• Apply basic filters using OpenCV.

3. Beginner-Friendly Explanation #

• Transformations: If you want to make the picture bigger (Scaling), you stretch the graph paper. Because you stretched it, you now have empty squares between your colors! The computer has to mathematically guess what color should fill the new empty squares.

• Filtering: Imagine looking at the graph paper through a tiny magnifying glass that only shows a 3x3 grid of squares. As you slide the magnifying glass across the paper, you mix the 9 colors together to create a new color. This "sliding window" is exactly how computers blur or sharpen an image!

4. Image Transformations (Affine) #

1. 1. Translation (Shifting): Shifting the entire image 50 pixels to the right.

1. 2. Rotation: Spinning the image by an angle (e.g., 45 degrees). When an image rotates, the corners might get cut off or create black triangles!

1. 3. Scaling (Resizing): Shrinking or enlarging. When enlarging, OpenCV uses *Interpolation*—a mathematical guess to create new pixels based on the colors of the surrounding original pixels.

5. What is a Kernel (Filter)? #

6. Blurring (Smoothing) #

• *Gaussian Blur:* A smarter, more natural-looking blur where the center pixel is given more mathematical "weight" than the outer pixels.

7. Sharpening #

8. Python Example: Applying Transformations #

import cv2

img = cv2.imread("street.jpg")
height, width = img.shape[:2]

# --- 1. ROTATION ---
# Calculate the center of the image to rotate around
center = (width/2, height/2)
# Create a rotation matrix: Center, Angle (45 deg), Scale (1.0 = original size)
rotation_matrix = cv2.getRotationMatrix2D(center, 45, 1.0)
# Apply the affine transformation
rotated_img = cv2.warpAffine(img, rotation_matrix, (width, height))


# --- 2. BLURRING ---
# Apply a Gaussian Blur using a 15x15 Kernel (Must be odd numbers!)
blurred_img = cv2.GaussianBlur(rotated_img, (15, 15), 0)

cv2.imwrite("rotated_blurred.jpg", blurred_img)

9. Mini Project #

10. Best Practices #

• Odd-Sized Kernels: When defining a Kernel for blurring in OpenCV (like (5, 5) or (11, 11)), the dimensions must always be odd numbers. This is because the sliding window needs a true "center" pixel to apply the calculation to!

11. Common Mistakes #

• Applying blur *after* edge detection: If you are trying to find the sharp edges of a car, you must apply the blur *first* to remove the static, and *then* run the edge detector. If you run the edge detector and then blur it, you destroy the edges you just found!

12. Exercises #

1. 1. Explain why enlarging an image (Scaling up) results in a loss of quality, while shrinking it does not.

13. Coding Challenges #

raw_image = load_image("4k_photo.jpg")

// Define the target dimensions required by the Neural Network
target_width = 224
target_height = 224

// Apply the scaling transformation
small_image = resize(raw_image, width=target_width, height=target_height)

save(small_image)

14. MCQs with Answers #

15. Interview Questions #

• Q: Explain the mathematical concept of Convolution (sliding a Kernel over an image) and how it achieves a blur effect.

• Q: What is Image Interpolation, and why is it necessary when scaling an image to a larger size?

16. FAQs #

17. Summary #

18. Next Chapter Recommendation #