Computer Vision - Part 2: Introduction
Computer Vision — Part 2: Introduction
Computer vision is the part of artificial intelligence that teaches computers to understand images and videos the way humans understand what they see.
When you look at a photo, you instantly recognise objects, their positions, depth, motion, and even mood. For a computer, the same photo is just a giant table of numbers. Computer vision is the collection of techniques that turn those numbers into useful meaning.
1. What is computer vision?
A simple definition:
Computer vision = giving a machine the ability to extract useful information from visual data.
That information can be:
| Task | Question the computer answers |
|---|---|
| Image classification | What is in this image? |
| Object detection | What objects are in the image and where are they? |
| Object localisation | Where is the single main object? |
| Image segmentation | Which pixels belong to each object? |
| Face recognition | Who is this person? |
| Optical character recognition (OCR) | What text is in this image? |
| Motion analysis | Which way are things moving? |
| Image generation | Can you create a new image from a description? |
2. Human vision vs computer vision
Your eyes are not cameras that simply record pixels. Your brain uses a lifetime of experience to recognise patterns, ignore clutter, and fill in missing information.
For example, you can recognise a dog in a photo even if:
- the dog is far away or very close
- the lighting is dim or very bright
- part of the dog is hidden behind a chair
- the photo is black-and-white
- the dog is upside down
A computer does none of this automatically. It sees a grid of colour values and must learn, from many examples, that a particular pattern of numbers means “dog”.
Human: "That is a cat."
Computer: "I see a 224 × 224 × 3 tensor of integers."
3. Why is computer vision difficult?
The same object can produce very different pixel patterns. The main challenges are:
| Challenge | What it means |
|---|---|
| Viewpoint changes | An object looks different from the front, side, or top. |
| Scale changes | A car can fill the image or be a tiny dot. |
| Illumination | Bright sunlight, shadows, and night photos change colours. |
| Occlusion | Objects hide behind each other. |
| Background clutter | The object blends into the background. |
| Deformation | A person can sit, run, or stretch. |
| Intra-class variation | Not all chairs, dogs, or cars look the same. |
A good computer vision system must be robust to all of these.
4. Images as numbers
Before any algorithm can work, an image must become numbers.
- A greyscale image is a 2-D grid of values. Each value is a pixel intensity, usually from 0 (black) to 255 (white).
- A colour image is a 3-D grid: height × width × channels, where the channels are red, green, and blue.
Grayscale: image[y, x] = one number (brightness)
Colour: image[y, x, c] = three numbers (R, G, B)
For a detailed look at pixels, colour models, and basic image arithmetic, see Part 1: Image Processing Fundamentals.
5. The computer vision pipeline
Most real-world vision systems follow the same broad steps:
camera, scanner,
video, medical device"] --> B["Preprocess
resize, crop, rotate,
denoise, normalise"] B --> C["Augment
flips, colour changes
(training only)"] C --> D["Model
extract features,
make a prediction"] D --> E["Postprocess
raw outputs →
human-readable results"] E --> F["Act
display, alert, store,
or control a system"] classDef step fill:#dbeafe,stroke:#1d4ed8,stroke-width:1.5px,color:#111 class A,B,C,D,E,F step
Preprocessing makes sure every input has the same size and range so the model sees consistent data. Augmentation artificially creates more training examples by slightly changing existing images. Postprocessing turns model outputs, such as tensors, into bounding boxes, masks, or labels.
6. The two big eras of computer vision
Era 1 — hand-crafted features
Before deep learning, engineers manually designed features such as:
- SIFT and SURF for finding interesting keypoints
- HOG (Histogram of Oriented Gradients) for describing object shapes
- Haar cascades for face detection
These methods worked well for specific, controlled problems, but they did not generalise. A feature designed for faces usually failed for cars.
Era 2 — learned features with deep learning
Convolutional neural networks (CNNs) learn their own features from data. Instead of a human writing rules, the network discovers edges, textures, shapes, and object parts by looking at thousands of examples.
Pixels → edges → textures → shapes → object parts → whole objects → prediction
This is the foundation of modern computer vision. The next part of this series, Part 3: CNNs for Computer Vision, covers just enough about how CNNs work to understand the detectors later in this series. For the full depth, formulas, and a real architecture walkthrough, see the Deep Learning section’s Part 9: Convolutional Neural Networks.
7. Common applications
Computer vision is used in many places you already interact with:
| Area | Example |
|---|---|
| Medical imaging | Detecting tumours in X-rays and MRI scans |
| Self-driving cars | Identifying pedestrians, lanes, traffic signs |
| Manufacturing | Finding defects on production lines |
| Retail | Self-checkout cameras and inventory counting |
| Security | Face unlock, intrusion detection |
| Agriculture | Crop-health monitoring from drone images |
| Entertainment | Snapchat filters, game motion capture |
| Accessibility | Describing photos for visually impaired users |
8. How computer vision relates to other topics
| Field | How it connects |
|---|---|
| Image processing | Low-level operations on pixels: filters, resizing, noise removal. |
| Deep learning | The main engine that learns features and predictions. |
| Natural language processing | Combining vision with text for image captioning or visual question answering. |
| Robotics | Cameras give robots information about the world around them. |
| Generative AI | Models that create or edit images, such as diffusion models and GANs. |
9. Summary
- Computer vision turns images and videos into useful information.
- The same object can look very different because of viewpoint, lighting, scale, and occlusion, which makes vision hard.
- A typical pipeline is: capture → preprocess → augment → model → postprocess → act.
- Modern computer vision is dominated by deep learning, especially CNNs.
- Vision connects to image processing, NLP, robotics, and generative AI.