Processing image data for Deep Learning

Neural network එකකට JPEG file එකක් කෙලින්ම කියවන්න බෑ. මේක තමයි ඒ පාලම — raw image files, model එකකට process කරන්න පුළුවන් numeric arrays බවට හරවනවා, tabular data එකට Pandas ඉටු කරපු ම role එකම.

A neural network can't read a JPEG file directly. This is the bridge — turning raw image files into numeric arrays a model can actually process, the same role Pandas played for tabular data.

21 May 20263 min readSelf Guide · Stage 17#Shipping Models to the World#DL 2. Image Data Processing#Machine learning

Lesson 62 of 64 · Part 5 · Shipping Models to the World · DL 2. Image Data Processing · bilingual notes (English + සිංහල)

Why this matters. A neural network can't read a JPEG file directly. This is the bridge — turning raw image files into numeric arrays a model can actually process, the same role Pandas played for tabular data.

සිංහලෙන්: Neural network එකකට JPEG file එකක් කෙලින්ම කියවන්න බෑ. මේක තමයි ඒ පාලම — raw image files, model එකකට process කරන්න පුළුවන් numeric arrays බවට හරවනවා, tabular data එකට Pandas ඉටු කරපු ම role එකම.

What you'll use this for. Every image-based project after this — including the Dog vs Cat classifier next — depends entirely on this exact pre-processing step being done correctly.

සිංහලෙන්: මින් පස්සේ එන image-based project ඕනම එකක් — ඊළඟට එන Dog vs Cat classifier එකත් ඇතුළුව — මුළුමනින්ම රඳා පවතින්නෙ මේ pre-processing step එක හරියට කරනවා කියන එක උඩ.

Key principles

Every image becomes a grid of numbers — an image is just a very structured matrix. A colour image is really three overlapping grids of pixel brightness values (red, green, blue) — resizing and normalizing these grids is what this lesson is about.
Image ඕනම එකක් numbers grid එකක් වෙනවා — image එකක් කියන්නෙ ඇත්තටම ගොඩක් structured matrix එකක් විතරයි

Walkthrough

Image data processing - conceptual video: https://youtu.be/Hs5RhjpQVRo

Topics:

Reading an image file and converting it to a numpy array
Resizing an image
RGB to Grayscale conversion

ExamplePython

# getting an image using web get

!wget 'https://tractive.com/blog/wp-content/uploads/2016/04/puppy-care-guide-for-new-parents.jpg'

Output

Output

--2022-03-17 15:37:15--  https://tractive.com/blog/wp-content/uploads/2016/04/puppy-care-guide-for-new-parents.jpg
Resolving tractive.com (tractive.com)... 54.192.18.89, 54.192.18.83, 54.192.18.105, ...
Connecting to tractive.com (tractive.com)|54.192.18.89|:443... connected.
HTTP request sent, awaiting response... 200 OK
Length: 113570 (111K) [image/jpeg]
Saving to: ‘puppy-care-guide-for-new-parents.jpg’

puppy-care-guide-fo 100%[===================>] 110.91K   382KB/s    in 0.3s    

2022-03-17 15:37:16 (382 KB/s) - ‘puppy-care-guide-for-new-parents.jpg’ saved [113570/113570]

Libraries that can be used for image processing:

matplotlib.image
Pillow
OpenCV (cv2)

ExamplePython

# importing the image module from matplotlib library

import matplotlib.image as mpimg
import matplotlib.pyplot as plt

ExamplePython

# loading an image through matplotlib.image module

img = mpimg.imread('/content/dog.jpg')

ExamplePython

type(img)

Output

Output

numpy.ndarray

ExamplePython

print(img.shape)

Output

Output

(1365, 2048, 3)

ExamplePython

print(img)

Output

Output

[[[147 182   0]
  [147 182   0]
  [147 182   0]
  ...
  [128 148   0]
  [128 148   0]
  [128 148   0]]

 [[146 181   0]
  [146 181   0]
  [147 182   0]
  ...
  [128 148   0]
  [128 148   0]
  [128 148   0]]

 [[145 180   0]
  [145 180   0]
… (31 more lines)

ExamplePython

# displaying the image from numpy array

img_plot = plt.imshow(img)
plt.show()

Processing image data for Deep Learning — figure 1

Resizing the image using Pillow library

ExamplePython

from PIL import Image

ExamplePython

img = Image.open('/content/dog.jpg')
img_resized = img.resize((200, 200))

ExamplePython

img_resized.save('dog_image_resized.jpg')

ExamplePython

# displaying the image from numpy array

img_res = mpimg.imread('/content/dog_image_resized.jpg')
img_res_plot = plt.imshow(img_res)
plt.show()

Processing image data for Deep Learning — figure 2

ExamplePython

print(img_res.shape)

Output

Output

(200, 200, 3)

Converting RGB images to Grayscale image using OpenCV

ExamplePython

# importing OpenCV library
import cv2

ExamplePython

img = cv2.imread('/content/dog.jpg')

ExamplePython

type(img)

Output

Output

numpy.ndarray

ExamplePython

img.shape

Output

Output

(1365, 2048, 3)

ExamplePython

grayscale_image = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)

ExamplePython

type(grayscale_image)

Output

Output

numpy.ndarray

ExamplePython

grayscale_image.shape

Output

Output

(1365, 2048)

cv2.imshow() will display the image. But this will not be allowed in Google Colab.

from google.colab.patches import cv2_imshow

ExamplePython

from google.colab.patches import cv2_imshow

ExamplePython

# displaying the image
cv2_imshow(grayscale_image)

Processing image data for Deep Learning — figure 3

ExamplePython

# saving the grayscale image
cv2.imwrite('dog_grayscale_image.jpg', grayscale_image)

Output

Output

True

Keep going: ← Lesson 61 · Lesson 63 → · Track it in the Self Guide

Finished this one?

Part of Self Guide · Stage 17: ML 5 — Shipping Models to the World. Progress is saved in this browser.

Guide

Previous Next

Projects in this lesson

ML & Vision

Machine Learning Playbook & Deep Learning Lab

Interactive machine learning educational playground and hands-on laboratory covering computer vision, regression, and deep neural nets.

Processing image data for Deep Learning

Key principles

Walkthrough

Image data processing - conceptual video: https://youtu.be/Hs5RhjpQVRo

Resizing the image using Pillow library

Machine Learning Playbook & Deep Learning Lab

Teaching Networking, DevOps and ML in Sinhala

Zero-Cost Hosting for InFocus SLUG 2025 — and 25 Hosting & Domain Options Compared

Life Vision: The Distributed Life