Morphological operations (erosion, dilation, opening, closing) in Computer Vision - ML Experiment: Train & Evaluate

Experiment - Morphological operations (erosion, dilation, opening, closing)

Problem:You have a binary image with noise and small holes in the objects. The current image processing pipeline uses erosion followed by dilation (opening) but the objects lose important details and the noise is not fully removed.

Current Metrics:Noise pixels remaining: 15%, Object detail loss: High

Issue:The current morphological operations remove noise but also erode important parts of the objects, causing loss of detail.

Your Task

Improve the morphological operations to reduce noise while preserving object details. Target: Noise pixels remaining < 5% and Object detail loss: Low.

You can only change the order and parameters of morphological operations (erosion, dilation, opening, closing).

Do not use other image processing techniques or filters.

Hint 1

Hint 2

Hint 3

Solution

Computer Vision

import cv2
import numpy as np

# Load binary image (0 background, 255 foreground)
image = cv2.imread('binary_image.png', cv2.IMREAD_GRAYSCALE)

# Define structuring element (kernel)
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3,3))

# Apply opening to remove noise (erosion followed by dilation)
opened = cv2.morphologyEx(image, cv2.MORPH_OPEN, kernel)

# Apply closing to fill small holes (dilation followed by erosion)
closed = cv2.morphologyEx(opened, cv2.MORPH_CLOSE, kernel)

# Save or display the result
cv2.imwrite('processed_image.png', closed)

# For evaluation, count noise pixels and check object details visually or with metrics

Added closing operation after opening to fill small holes inside objects.

Used a 3x3 rectangular structuring element to balance noise removal and detail preservation.

Combined opening and closing instead of only opening.

Results Interpretation

Before: Noise pixels remaining: 15%, Object detail loss: High

After: Noise pixels remaining: 3%, Object detail loss: Low

Using a combination of opening and closing morphological operations helps remove noise and fill small holes without losing important object details.

Bonus Experiment

Try using different shapes and sizes of structuring elements (e.g., ellipse, cross, larger sizes) to see how they affect noise removal and detail preservation.

💡 Hint

Larger kernels remove more noise but may erode details; different shapes affect how edges are processed.

Practice

1. What does the erosion operation do to the white parts of a binary image?

easy

A. It grows the white parts by adding pixels at the edges.

B. It removes noise by smoothing the edges.

C. It fills small holes inside the white parts.

D. It shrinks the white parts by removing pixels at the edges.

Solution

1. Step 1: Understand erosion effect on white pixels

   Erosion removes pixels from the boundaries of white regions, making them smaller.

2. Step 2: Compare with other operations

   Dilation grows white parts, opening removes noise, closing fills holes, so erosion must shrink white parts.

3. Final Answer:

   It shrinks the white parts by removing pixels at the edges. -> Option D

4. Quick Check:

   Erosion = Shrinks white parts [OK]

Hint: Erosion shrinks white areas by cutting edges [OK]

Common Mistakes:

• Confusing erosion with dilation
• Thinking erosion fills holes
• Mixing erosion with noise removal

2. Which of the following is the correct syntax to perform dilation using OpenCV in Python on an image img with a 3x3 kernel?

easy

A. cv2.erode(img, np.ones((3,3), np.uint8))

B. cv2.dilate(img, np.ones((3,3), np.uint8))

C. cv2.dilate(img, (3,3))

D. cv2.dilate(img, kernel=3)

Solution

1. Step 1: Recall dilation syntax in OpenCV

   Dilation requires the image and a structuring element (kernel), usually created with np.ones of desired size and type.

2. Step 2: Check options for correct usage

   cv2.dilate(img, np.ones((3,3), np.uint8)) uses cv2.dilate with a 3x3 kernel created by np.ones and correct dtype, which is valid syntax.

3. Final Answer:

   cv2.dilate(img, np.ones((3,3), np.uint8)) -> Option B

4. Quick Check:

   Dilation syntax = cv2.dilate(image, kernel) [OK]

Hint: Use np.ones((3,3), np.uint8) as kernel for dilation [OK]

Common Mistakes:

• Using erode instead of dilate
• Passing kernel size tuple directly
• Using wrong kernel datatype

3. Given the following Python code using OpenCV:

import cv2
import numpy as np
img = np.array([[0,0,0,0,0],
                [0,255,255,255,0],
                [0,255,0,255,0],
                [0,255,255,255,0],
                [0,0,0,0,0]], dtype=np.uint8)
kernel = np.ones((3,3), np.uint8)
eroded = cv2.erode(img, kernel)
print(eroded)

What will be the printed output?

medium

A. [[ 0 0 0 0 0]
[ 0 255 0 255 0]
[ 0 0 0 0 0]
[ 0 255 0 255 0]
[ 0 0 0 0 0]]

B. [[ 0 0 0 0 0]
[ 0 255 255 255 0]
[ 0 255 255 255 0]
[ 0 255 255 255 0]
[ 0 0 0 0 0]]

C. [[ 0 0 0 0 0]
[ 0 0 0 0 0]
[ 0 0 0 0 0]
[ 0 0 0 0 0]
[ 0 0 0 0 0]]

D. [[255 255 255 255 255]
[255 255 255 255 255]
[255 255 255 255 255]
[255 255 255 255 255]
[255 255 255 255 255]]

Solution

1. Step 1: Understand erosion on the given image

   Erosion removes pixels at edges of white regions. The center pixel (0) surrounded by 255s will cause erosion to shrink the white area.

2. Step 2: Apply 3x3 kernel erosion

   Since the kernel covers neighbors, any pixel with a zero neighbor becomes zero. The white cross shape will erode to a smaller cross with zeros at the center and edges.

3. Final Answer:

   White cross with zeros at center and edges as shown in option A -> Option A

4. Quick Check:

   Erosion shrinks white, so edge pixels vanish but some inner pixels remain [OK]

Hint: Erosion removes edge pixels, shrinking white areas [OK]

Common Mistakes:

• Assuming erosion keeps center pixels
• Confusing erosion with dilation output
• Ignoring zero pixels in kernel neighborhood

4. You wrote this code to perform opening on an image img but it does not remove noise as expected:

kernel = np.ones((3,3), np.uint8)
opened = cv2.dilate(cv2.erode(img, kernel), kernel)

What is the error and how to fix it?

medium

A. Kernel size is too small; increase kernel size to remove noise.

B. The order is reversed; opening is dilation followed by erosion, so swap the calls.

C. Use cv2.morphologyEx with cv2.MORPH_OPEN instead for correct opening.

D. The order is reversed; opening is erosion followed by dilation, so code is correct.

Solution

1. Step 1: Check the definition of opening

   Opening is erosion followed by dilation. The code applies erosion then dilation, which is correct in order.

2. Step 2: Identify practical issue

   Manual calls may work but can be error-prone; using cv2.morphologyEx with MORPH_OPEN is recommended for correct and optimized opening.

3. Final Answer:

   Use cv2.morphologyEx with cv2.MORPH_OPEN instead for correct opening. -> Option C

4. Quick Check:

   Use built-in morphologyEx for opening [OK]

Hint: Use cv2.morphologyEx with MORPH_OPEN for opening [OK]

Common Mistakes:

• Swapping erosion and dilation order
• Not using built-in morphology functions
• Assuming kernel size fixes logic errors

5. You have a noisy binary image with small black holes inside white objects. Which morphological operation should you apply to fill these holes without changing the object shapes much?

hard

A. Closing

B. Dilation

C. Opening

D. Erosion

Solution

1. Step 1: Understand the problem of black holes inside white objects

   Black holes are small dark spots inside white regions that need to be filled.

2. Step 2: Choose operation that fills holes without shrinking objects

   Closing is dilation followed by erosion; it fills small holes and gaps while preserving object shape.

3. Final Answer:

   Closing -> Option A

4. Quick Check:

   Closing fills holes inside white objects [OK]

Hint: Closing fills holes inside white objects [OK]

Common Mistakes:

• Using erosion which shrinks objects
• Using opening which removes noise but not holes
• Confusing dilation alone with closing