Computer vision is the field of AI that gives machines the ability to "see": to extract meaning from images, video, or sensor streams. To a computer, a photo is just a grid of numbers (pixels); the challenge is turning that raw matrix into useful concepts — "this is a face," "this tumor is 8 mm," "this pedestrian is crossing."
From pixels to meaning
A color image is a tensor of shape height × width × 3 (red, green, blue channels). Each pixel ranges from 0 to 255:
$$ I \in {0, 1, \dots, 255}^{H \times W \times 3} $$
Historically, engineers hand-crafted feature descriptors (edges, textures, corners). Since 2012, Convolutional Neural Networks (CNNs) and more recently Vision Transformers (ViTs) learn the relevant features by themselves from millions of labeled examples.
The core tasks
| Task | Question asked | Example |
|---|---|---|
| Classification | What is it? | cat / dog |
| Object detection | Where, and how many? | boxes around cars |
| Segmentation | Which exact pixels? | precise outline of an organ |
| Pose estimation | What posture? | joints of a body |
Real-world applications
- Healthcare: spotting lesions on X-rays and MRIs.
- Mobility: perception for autonomous vehicles.
- Industry: automated quality control on production lines.
- Agriculture: drones detecting crop disease and water stress.
Computer vision does not "see" the way we do: it computes statistical correlations. Its power — and its biases — depend entirely on the data it was trained on.