Basic Image Processing Using NumPy

Quick answer: Learn how to represent and manipulate images as NumPy arrays, including reading pixel values, cropping, flipping, and basic filtering.

Images as NumPy Arrays

A digital image is fundamentally just a grid of numbers. A grayscale image is a 2D array of pixel intensities, and a colour image is a 3D array with an extra dimension for the red, green and blue channels.

import numpy as np
from PIL import Image

image = Image.open('photo.jpg')
img_array = np.array(image)

print(img_array.shape)   # e.g. (400, 600, 3) -- height, width, channels

Reading and Modifying Pixel Values

print(img_array[0, 0])          # RGB values of the top-left pixel

img_array[0:50, 0:50] = [255, 0, 0]   # paints a red square in the top-left corner

Cropping an Image

cropped = img_array[100:300, 150:400]   # rows 100-300, columns 150-400

Since the image is just an array, cropping is simply array slicing, exactly like slicing a list or a Pandas DataFrame.

Flipping an Image

flipped_horizontal = img_array[:, ::-1]   # reverse columns
flipped_vertical = img_array[::-1, :]     # reverse rows

Converting to Grayscale

grayscale = np.mean(img_array, axis=2).astype(np.uint8)

Averaging the three colour channels gives a simple, if not perceptually accurate, grayscale conversion. Libraries like OpenCV use a weighted formula that better matches human colour perception.

Simple Brightness and Contrast Adjustment

brighter = np.clip(img_array.astype(int) + 40, 0, 255).astype(np.uint8)

np.clip() is essential here, since adding 40 to a pixel already near 255 would otherwise overflow and wrap around incorrectly in an 8-bit integer type.

Basic Image Statistics

print(img_array.mean())    # average pixel intensity across the image
print(img_array.std())     # spread of pixel intensities
print(img_array.max(), img_array.min())

Why This Matters for Machine Learning

Understanding that images are just numeric arrays is the foundation for everything that follows, from basic preprocessing steps like normalisation and resizing, to feeding image data into a Convolutional Neural Network, which operates directly on these same underlying arrays.

Common Interview Questions

Why does adding a brightness value need np.clip()?

Pixel values in an 8-bit image are stored as integers from 0 to 255. Without clipping, adding a value could push a pixel above 255, causing it to wrap around incorrectly rather than simply capping at the maximum brightness.

How would you crop an image using NumPy?

Since the image is represented as an array, cropping is simply slicing that array by row and column range, exactly like slicing a list.