What Is Computer Vision? What Software Can Read From Images
Computer vision is how software extracts meaning from images. This guide walks through the eleven features of a real vision service and which ones suit a small project.
Computer vision is the part of AI that lets software get meaning from images and video. A phone that recognises a face to unlock, a store app that reads a barcode, and a tool that tells you a photo contains a dog are all computer vision. It is older than the current wave of chatbots, and it still does many jobs that chatbots do badly or expensively. If you are adding anything to an app that looks at pictures, you will end up choosing between a dedicated vision service and a general AI model.
What computer vision can do
The clearest way to see the range is to read the feature list of a real service. Google's Cloud Vision documentation lists eleven things its API can do with a single image:
Feature | What it returns |
|---|---|
Label detection | Generalised labels for the image, each with a confidence score |
Object localization | Labels and bounding boxes for multiple objects in one image |
Text detection | Optical character recognition for an image |
Document text detection | OCR tuned for PDFs, dense pages and handwriting |
Face detection | Where faces are, plus facial landmarks and image conditions such as blur |
Landmark detection | The name of a landmark, with a confidence score |
Logo detection | A logo's description, confidence score and bounding polygon |
Image properties | Dominant colours in the image |
Crop hints | A suggested crop region for chosen aspect ratios |
Web entities and pages | Related web content, matching images and a best-guess label |
Explicit content detection | The likelihood of categories such as adult content or violence |
The same page notes that face detection locates faces but does not support recognising individuals. You can read the full feature list in Google's documentation.
How it works, in one paragraph
Vision systems are trained on large collections of labelled images until they learn patterns, such as the arrangement of edges and textures that signals a cat. They then apply those patterns to new images and report what they find, usually with a confidence score. A score of 0.9 means the system is fairly sure, not that it is right. That is why important decisions should not rest on a single detection.
Computer vision versus a general AI model
A vision-language model accepts an image and a question in plain language, and answers in text. A vision API returns structured results from a fixed menu. What a vision-language model is covers the first kind in detail.
Choose a vision API for narrow, repeatable work: reading text, tagging thousands of photos, detecting unsafe uploads. Output is predictable, and pricing is per image, listed in the provider's documentation.
Choose a general model for open questions about an image, like "does this shelf look tidy?" or "what is wrong with this screenshot?" Output is flexible and less predictable.
Many products use both, with the vision API doing the cheap first pass.
Ideas that suit a small project
Receipts and invoices: read the text and pull out totals. What OCR is and choosing an AI model for document extraction cover this.
Photo upload moderation: flag explicit images before they appear on your site.
Product photo tidy-up: use object boxes and crop hints to frame images consistently.
Alt text drafts: label detection gives a starting point for image descriptions, which a person should still review.
Inventory counts from a photo: object localization can count items on a shelf, though accuracy depends on the setup.
What to be careful about
Images often contain people, and that brings obligations that text does not.
Faces and biometrics. Detecting that a face exists is different from identifying a person. Rules on biometric data vary by country, so check before storing anything about faces.
Bias and error. Vision systems perform unevenly across lighting, skin tones, camera types and cultures. Test on your own real images.
Hidden instructions. Text inside an image can be aimed at an AI model downstream. Prompt injection through an image explains the risk.
Confidence is not truth. Set a threshold, and route low-confidence results to a person.
For the broader picture of how these systems relate to chat models, how AI models work is the best place to continue.
FAQ
Is computer vision the same as AI?
It is a branch of AI focused on images and video. It includes older techniques as well as modern neural networks.
What is the difference between computer vision and image recognition?
Image recognition usually means naming what is in an image. Computer vision is the wider field, including reading text, locating objects and finding faces.
Can computer vision identify people?
Some systems can, but not all. Google's Vision API, for one, detects faces without supporting individual recognition. Identifying people carries legal and ethical obligations.
Do I need computer vision to read text in images?
OCR is the specific tool for that, and it is one feature inside computer vision.
How did this land?
About the author

Senior Editor, AI & Product
Cecilia leads the Swarmz editorial desk. She has spent a decade turning complex AI and product topics into writing people actually finish, and she owns the blog's quality bar.


