What is computer vision?
In one line
Ages 8 to 10: A computer sees a picture as lots of numbers, and computer vision finds patterns in those numbers to say what is in it.
Ages 11 to 13: Every pixel is stored as numbers for red, green and blue, and a model learns from labeled pictures which number patterns go with which answer.
Ages 14 to 18: Modern vision systems are usually deep neural networks trained on labeled images, and they must be tested on fresh images because patterns that fit the training pictures may not hold on new ones.
Computer vision is the field that teaches computers to make sense of pictures and video. To a computer, a photo is a grid of numbers, one set for each tiny dot of color. Computer vision turns those numbers into useful answers, such as "this is a cat" or "a person is crossing".
Last reviewed: 20 October 2026 7 min read
The idea in one picture
Imagine a mosaic made of tiles, where every tile has a number painted on it instead of a color. A computer gets a picture in exactly that form. Each tile is a pixel, and in a black-and-white image each pixel is a number from 0 (black) to 255 (white).
Here is a tiny 4 × 4 picture of the letter L:
| 255 | 0 | 0 | 0 |
|---|---|---|---|
| 255 | 0 | 0 | 0 |
| 255 | 0 | 0 | 0 |
| 255 | 255 | 255 | 0 |
The computer never sees an "L". It sees sixteen numbers. Computer vision is the work of getting from those numbers to the answer "L". A color picture stores three numbers per pixel, one each for red, green and blue, so even a small 32 × 32 color image is 32 × 32 × 3 = 3,072 numbers.
How it works, step by step
- Capture. A camera turns light into pixels, and each pixel is stored as numbers.
- Collect examples. People gather many pictures and label each one with the right answer, such as "cat" or "not cat".
- Learn. A model studies the labeled pictures and adjusts itself until the number patterns it finds match the labels.
- Predict. Given a new picture, the model gives its best answer.
- Test. The model is checked on pictures it has never seen, and every mistake is counted.
A little deeper (ages 11 to 13)
The same cat can produce very different numbers. A cat in dim light, seen from the side, half behind a chair, gives a completely different grid from the same cat in sunshine. Textbooks list these as the main challenges: viewpoint, size, lighting, objects hiding other objects, busy backgrounds, and the huge variety inside one category, like all the different shapes of chair.
Computer vision covers several kinds of task. Classification gives the whole picture one label. Detection finds each object and draws a box around it. Segmentation labels every pixel, for example "road" or "sky".
Going further (ages 14 to 18)
Most modern vision systems are deep neural networks, often convolutional networks, which look at small patches of pixels and build up larger patterns layer by layer.
When researchers inspect trained networks, the first layer is usually the easiest to interpret, while some deep units respond to whole things such as text or upper bodies.
A vision model is only as good as its examples and tests. If some groups of people are under-represented in the training pictures, the model can be less accurate for those groups. One well-known study of commercial systems that guess gender from a face photo found that darker-skinned women were misclassified more often than any other group it tested. That is why good practice reports accuracy separately for different groups, not only one overall score.
Common mix-ups
- Mix-up: "The computer sees the picture like I do."Fact: It receives numbers and finds patterns that it learned from examples. It has no idea what a cat is beyond those patterns.
- Mix-up: "Computer vision just means face recognition."Fact: Faces are one use. Others include robots, cars that sense the road and apps that recommend pictures or videos.
- Mix-up: "The camera does the seeing."Fact: A camera only captures pixels. Computer vision is the part that turns pixels into an answer.
- Mix-up: "It was right on its training pictures, so it works."Fact: Only a test on new pictures shows that.
Try it (no device, 5 minutes)
Draw two empty 5 × 5 grids. One person secretly shades a simple shape in their grid, then reads it out row by row as numbers: 1 for shaded, 0 for empty. The other person fills in their grid from the numbers alone and guesses the shape. Then move the shape one square to the right and read it out again. Several numbers change, yet it is the same shape. That is one reason computer vision is hard.
Quick check
How does a computer store a black-and-white picture?
Answer: As a grid of numbers, one for each pixel. In the usual format, 0 is black, 255 is white and the numbers in between are grays.
As a grid of numbers, one for each pixel. In the usual format, 0 is black, 255 is white and the numbers in between are grays.A model gets every one of its training pictures right. Do you know it works?
Answer: Not yet. You need to test it on pictures it has never seen and count the mistakes.
Not yet. You need to test it on pictures it has never seen and count the mistakes.A model learned only from daytime photos. Why might it fail at night?
Answer: Lighting changes the pixel numbers a lot, and the model never learned which nighttime patterns go with which answer.
Lighting changes the pixel numbers a lot, and the model never learned which nighttime patterns go with which answer.Questions
Computer vision is the field that teaches computers to make sense of pictures and video. The computer receives each picture as a grid of numbers and learns which number patterns go with answers such as "cat" or "road".
People collect many pictures labeled with the right answers, and a model learns which patterns of pixel numbers match each label. The model then gives its best answer for new pictures, and it is tested on pictures it has never seen.
Robots and cars that sense what is around them use computer vision. So do apps that sort photos or recommend pictures and videos, and systems that find and box objects in an image.
Computer vision is one field within AI. Most modern computer vision uses machine learning, especially deep neural networks, to learn from labeled pictures.
Yes. Changes in lighting, angle or background can change the pixel numbers enough to fool a model, and a model can be less accurate for groups that were missing from its training pictures. Testing on new pictures and on different groups shows where it fails.
Mechna Academy is a city of machines for curious children aged 8 to 18, opening first to ages 8 to 10. In each mission, a child guesses how a machine decides, then opens its Peek Hatch to check, changes one thing and tests it on a case it has never seen. Children learn how AI works this way, and they never chat with an AI.
For grown-ups: join the waitlistPlaces are limited. The waitlist is for parents and guardians 18 and older.