Machine Learning · Chapter 38 of 40
Image Basics
An image is a grid of PIXELS. Colour images have 3 CHANNELS (RGB). Classic ML uses hand-crafted features (edges, histograms); deep learning uses CNNs (Convolutional Neural Networks).
CNNs learn features automatically.
Example 1 (python)
import numpy as np
img = np.random.rand(28, 28, 3) # HxWxChannels
print(img.shape)Output
(28, 28, 3)Represent as a 3D array.
Example 2 (python)
# Simple CNN with Keras
# tf.keras.layers.Conv2D(32, (3,3), activation='relu')Convolution filters detect local patterns.
Key points
- Images are grids of pixels.
- Colour has 3 channels (RGB).
- CNNs learn image features automatically.
- Always normalize pixel values (0–1).
💡 Note: Always divide pixel values by 255 to scale them to [0, 1] before feeding a neural net.