4.2 How Computers Represent Images

What is image?

If you want to understand how computers recognize and interpret images, you first need to know what an image actually is from a computer's point of view.

To humans, an image may look like a meaningful scene filled with objects, colors, and emotions. But to a computer, an image is simply a grid of numerical data made up of pixels, where each pixel stores values that represent brightness or color. By processing these numbers with mathematical models and algorithms, computers can begin to detect patterns, identify shapes, distinguish objects, and ultimately make decisions about what an image contains. Understanding this difference between human perception and machine representation is the first step toward understanding how computer vision works.

Images as Arrays of Numbers

A digital image is a grid of values. Each location in the grid is called a pixel.

• In a grayscale image, each pixel usually stores one value.
• In a color image, each pixel stores multiple values, one for each channel.

For a computer, an image is not "a cat" or "a car." It is an array with shape, datatype, and numeric content.

Channels

A channel is one kind of value stored at each pixel.For example, a grayscale image has 1 channel, while a color image often has 3 channels such as RGB. In other words, a computer sees an image as numbers arranged by channel, not directly as real-world objects. OpenCV also often uses BGR order instead of RGB, Many beginners first meet images as RGB images. But different tasks use different channel layouts.

OpenCV commonly uses BGR order instead of RGB, which is an important practical detail.

Resolution

Resolution describes how many pixels the image contains, such as 640 x 480 or 1920 x 1080.

Higher resolution usually gives more detail, but it also increases:

• memory use
• processing cost
• training and inference time

Video as a Sequence of Frames

A video is a time-ordered stream of images.

That means video introduces two layers of complexity:

• per-frame visual structure
• temporal change across frames

Challenge

Real camera input can be affected by:

• motion blur
• low light
• noise
• lens distortion
• compression artifacts
• dropped frames

This is why good computer vision work depends not only on models, but also on data quality.

Have a try

Operate an Image with OpenCV

#clone the source code
git clone https://github.com/Seeed-Projects/reComputer-Jetson-for-Beginners
cd reComputer-Jetson-for-Beginners/4-Computer-Vision/4.2-How-Computers-Represent-Images/code
# Install opencv if you don't install before
pip install opencv-python
#run the script
python show_img.py

You will see the output:

Reading Video

python show_video.py

You will see the output:

Exercises / Reflection

1. Load one image and print its shape and dtype.
2. Convert the same image into grayscale and HSV. Describe what changes visually.
3. Resize one image to two different resolutions and compare file size or processing time.
4. Open a short video and count approximately how many frames are displayed in 10 seconds.

Summary

Images are structured numeric data, not just pictures. Understanding pixels, channels, resolution, and video frames is essential because every later algorithm, from thresholding to CNNs, operates on these representations.

Suggested Next Step

Continue to 4.3 Classical Computer Vision.

References

• OpenCV Documentation
• OpenCV Python Tutorials