10.1.3 OpenCV Basics

Learning Objectives

After completing this section, you will be able to:

• Use OpenCV to create, read, and save images
• Perform basic transformations such as resizing, cropping, and flipping
• Understand common color order issues in OpenCV
• Draw rectangles, circles, and text on images with OpenCV

---

Why do almost every CV beginner course start with OpenCV?

Because OpenCV is like the “Swiss Army knife” of computer vision:

• It can read and write images
• It can resize, rotate, and crop
• It can do filtering and edge detection
• It can do face detection and video processing

And it is very suitable for beginners to build an engineering mindset.

Install Dependencies

The following code can run directly:

pip install opencv-python numpy

---

First create an image instead of relying on an external file

To make the code run directly, let’s generate a blank image ourselves first.

import cv2
import numpy as np

# Create a black canvas: height 240, width 320, 3 color channels
img = np.zeros((240, 320, 3), dtype=np.uint8)

print("shape:", img.shape)
print("dtype:", img.dtype)

cv2.imwrite("opencv_blank.png", img)
print("Saved opencv_blank.png")

Expected output:

shape: (240, 320, 3)
dtype: uint8
Saved opencv_blank.png

Here, shape = (240, 320, 3) means:

• Height: 240
• Width: 320
• 3 color channels

---

The color order in OpenCV is BGR, not RGB

This is a very classic pitfall.

OpenCV uses:

BGR

by default, not the RGB we are more familiar with.

import cv2
import numpy as np

img = np.zeros((100, 100, 3), dtype=np.uint8)

# This color is BGR, not RGB
img[:, :] = (255, 0, 0)

cv2.imwrite("opencv_blue.png", img)
print("Saved a blue image opencv_blue.png")

Expected output:

Saved a blue image opencv_blue.png

If you think (255, 0, 0) is red, you will end up with a “wrong color” image.

Convert to RGB

import cv2
import numpy as np

img_bgr = np.zeros((2, 2, 3), dtype=np.uint8)
img_bgr[:, :] = (255, 0, 0)

img_rgb = cv2.cvtColor(img_bgr, cv2.COLOR_BGR2RGB)

print("BGR pixel:", img_bgr[0, 0].tolist())
print("RGB pixel:", img_rgb[0, 0].tolist())

Expected output:

BGR pixel: [255, 0, 0]
RGB pixel: [0, 0, 255]

---

Evidence to Keep

Keep this page’s proof of learning as a small evidence card:

Input Image

source image or synthetic image used in the run

Array Shape

width, height, channels, dtype, and coordinate convention

Processed Output

grayscale, crop, edge, threshold, or saved intermediate image

Failure Check

channel order, resize distortion, coordinate mistake, or over-processing

Expected Output

before/after image plus the printed shape or pixel values

Common basic operations: resizing, cropping, flipping

import cv2
import numpy as np

img = np.zeros((200, 300, 3), dtype=np.uint8)
img[:, :] = (40, 180, 240)

# Resize
small = cv2.resize(img, (150, 100))

# Crop: rows first, then columns, i.e. [y1:y2, x1:x2]
crop = img[50:150, 80:220]

# Flip
flip_horizontal = cv2.flip(img, 1)

print("Original image:", img.shape)
print("After resizing:", small.shape)
print("After cropping:", crop.shape)
print("After horizontal flip:", flip_horizontal.shape)

cv2.imwrite("opencv_small.png", small)
cv2.imwrite("opencv_crop.png", crop)
cv2.imwrite("opencv_flip.png", flip_horizontal)

Expected output:

Original Image

(200, 300, 3)

After Resizing

(100, 150, 3)

After Cropping

(100, 140, 3)

After Horizontal Flip

(200, 300, 3)

Why is cropping written as [y1:y2, x1:x2]?

Because an image is essentially a 2D array, and array indexing follows this order:

1. Rows first (height direction, y)
2. Then columns (width direction, x)

Reading Tip

Two of the most common beginner mistakes in OpenCV are: the default color order is BGR, not RGB; and array cropping uses y before x. When reading this diagram, think of the image as a “row-column array” rather than a flat coordinate plane.

---

Drawing on images

Many computer vision tasks need results marked on the image, such as:

• Drawing bounding boxes
• Labeling class names
• Marking center points

import cv2
import numpy as np

canvas = np.ones((300, 400, 3), dtype=np.uint8) * 255

# Draw rectangle
cv2.rectangle(canvas, (50, 50), (180, 180), (0, 255, 0), 2)

# Draw circle
cv2.circle(canvas, (280, 120), 40, (255, 0, 0), -1)

# Draw line
cv2.line(canvas, (30, 250), (350, 250), (0, 0, 255), 3)

# Write text
cv2.putText(
    canvas,
    "CV Demo",
    (120, 40),
    cv2.FONT_HERSHEY_SIMPLEX,
    1,
    (0, 0, 0),
    2
)

cv2.imwrite("opencv_draw_demo.png", canvas)
print("Saved opencv_draw_demo.png")

Expected output:

Saved opencv_draw_demo.png

---

Convert to a grayscale image

Many classic vision operations first convert a color image to grayscale because:

• It is faster to compute
• It removes color distractions
• It keeps only brightness information

import cv2
import numpy as np

img = np.zeros((100, 100, 3), dtype=np.uint8)
img[:, :50] = (0, 0, 255)      # Red
img[:, 50:] = (0, 255, 0)      # Green

gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

print("Original image shape:", img.shape)
print("Grayscale image shape:", gray.shape)
print("First 5 pixels of grayscale image:", gray[0, :5].tolist())

cv2.imwrite("opencv_gray.png", gray)

Expected output:

Original image shape: (100, 100, 3)
Grayscale image shape: (100, 100)
First 5 pixels of grayscale image: [76, 76, 76, 76, 76]

---

A small project: make an “info card” image

This example combines the knowledge from above: creating an image, drawing shapes, writing text, and saving the result.

import cv2
import numpy as np

card = np.ones((220, 420, 3), dtype=np.uint8) * 245

cv2.rectangle(card, (20, 20), (400, 200), (60, 120, 200), 2)
cv2.circle(card, (80, 85), 35, (60, 120, 200), -1)

cv2.putText(card, "AI Fullstack", (140, 75), cv2.FONT_HERSHEY_SIMPLEX, 0.9, (30, 30, 30), 2)
cv2.putText(card, "Chapter 10: CV Basics", (140, 115), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (60, 60, 60), 2)
cv2.putText(card, "OpenCV starter demo", (40, 170), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (20, 20, 20), 2)

cv2.imwrite("opencv_info_card.png", card)
print("Saved opencv_info_card.png")

Expected output:

Saved opencv_info_card.png

---

Common beginner mistakes

cv2.imshow() does not open a window

In many remote environments, notebooks, and server environments, imshow() is not convenient to use. For teaching and script-based scenarios, it is recommended to use cv2.imwrite() to save the result first.

Treating BGR as RGB

This is one of the most common bugs for OpenCV beginners.

Reversing the x and y order when cropping

Image array indexing is [y, x], not [x, y].

---

Summary

The key point of this lesson is not to memorize every OpenCV API, but to build the feeling that “I can already manipulate images”:

• I can create images
• I can transform images
• I can annotate images
• I can save the results

With these basics, the next lesson on filtering, edge detection, and morphological operations will be much smoother.

---

Exercises

1. Change the canvas color to another color and generate a new card image.
2. Draw multiple rectangles and circles on the same image to practice the coordinate system.
3. Try resizing the image to different resolutions and then save the results.

Solution approach and explanation

1. If you use OpenCV drawing functions, remember that color tuples are usually BGR, not RGB. A correct new card should save successfully and show the intended color after you open it.
2. For rectangles and circles, check that all coordinates stay inside the image. Drawing order matters: later shapes can cover earlier ones.
3. Resizing changes the number of pixels. If the width-height ratio changes, the image is distorted, so keep both an intentionally distorted version and an aspect-ratio-preserving version when comparing.