Stop 10 · Computer VisionTopics · 2/9

OpenCV Basics

In short

Before any deep learning, someone has to read the image, crop it, blur it, find its edges or grab frames from a webcam. OpenCV does all of that, fast, in a few lines.

The standard library for classical image processing and computer vision operations — reading, transforming, and analyzing images and video outside of deep learning too.

The blob trims and tints a flower photo with scissors and filter sheets at a photocopy shop counter.

Think of it like this

The photocopy shop toolkit

The xerox shop near college has a paper cutter, a laminator, colour settings and a scanner. Bring any document and they can crop, enlarge, darken or copy it. OpenCV is that shop for images: no AI needed, just very handy tools.

Watch it happen

Finding edges in a photo

imread

Grayscale

Blur

Canny

imwrite

1/5cv2.imread gives a NumPy array of shape (height, width, 3) in BGR order.

The big ideas

What you need to know

Images are NumPy arrays

An OpenCV image is just an array. Slicing img[100:300, 50:250] crops it. No special image object needed.

BGR by default

OpenCV stores colour as blue, green, red. Convert before showing with Matplotlib or sending to most models.

Classic operations

Resize, rotate, threshold, blur, edge detection, contours and drawing boxes. Most CV projects use a few of these.

Video is many frames

cv2.VideoCapture reads a webcam or file one frame at a time, so the same image code works on video.

See it in code

Walk through the code

python

1# pip install opencv-python2import cv23 4img = cv2.imread("auto.jpg")          # BGR array, None if path is wrong5print(img.shape)                      # e.g. (720, 1280, 3)6 7gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)8blur = cv2.GaussianBlur(gray, (5, 5), 0)9edges = cv2.Canny(blur, 50, 150)10 11crop = img[100:400, 200:600]          # rows first, then columns12cv2.rectangle(img, (200, 100), (600, 400), (0, 255, 0), 2)13cv2.imwrite("edges.jpg", edges)

1/3imread returns None instead of an error when the file is missing, so check the shape.

Read, convert, blur, detect edges, crop and draw a box. Note that slicing uses (y, x) but drawing uses (x, y).

Save yourself the pain

Common mistakes

Oops

Showing an OpenCV image with plt.imshow and wondering why the colours look strange.

Do this instead

Convert with cv2.cvtColor(img, cv2.COLOR_BGR2RGB) before plotting, or use cv2.imshow outside notebooks.

Oops

Mixing up (x, y) and (row, col).

Do this instead

Array indexing is img[y, x]. Functions like cv2.rectangle and cv2.resize take (x, y) or (width, height).

Your mission

Make a sketch filter

~30 min

Quick check

You want the region from x=50 to x=150 and y=20 to y=80. Which slice is correct?

Go deeper

1Practice reading, displaying and transforming images with OpenCV.
2Learn basic operations: edge detection, thresholding, color space conversion.
3Combine OpenCV preprocessing with a deep learning model in one pipeline.