AI glossary for businesses
Computer vision
In short
Computer vision is the area of AI that lets software interpret images and video: recognising objects, reading text, counting items or spotting defects. It sits behind face recognition on phones, document scanning and automatic photo tagging. For businesses, it turns pictures, scans and camera feeds into data you can search and use.
How computer vision works
To software, a photo is just a grid of coloured dots. A computer vision model learns, from millions of labelled examples, which patterns of dots usually mean "shoe", "invoice total" or "scratch on the surface". Once trained, it can look at a new image and say what is in it, where it is and how sure it is.
A good analogy is a new warehouse worker. On day one, every box looks the same. After a few weeks of seeing labels, shapes and colours, they recognise products at a glance. A vision model goes through the same learning curve, only with far more examples.
Common tasks include image classification (what is this?), object detection (where is it?), optical character recognition or OCR (what does the text say?) and quality inspection (is something wrong?). Many newer models combine vision with language. That is why you can now upload a photo to ChatGPT or Google Gemini and ask questions about it. This combination is called multimodal AI.
In practice for a small business
A small accounting office in Patras receives supplier invoices as photos and PDFs. OCR plus a vision model can read the supplier, VAT number, date and amount, then fill a spreadsheet for a person to check. The use case on invoice and document data extraction shows how to set this up.
In retail, an e-shop can use vision to tag product photos by colour and style, or to flag images with a messy background before they go live. A small manufacturer can photograph finished parts and let a model highlight possible defects for a person to confirm.
Why it matters and what to watch out for
Image quality decides results. Blurry phone photos, poor lighting and handwriting lower accuracy. Test with your real documents, not clean samples.
Keep a human check. A misread digit on an invoice is a real cost. Plan for review, at least at the start.
Faces and people are sensitive. Camera footage of customers or staff is personal data under GDPR. Facial recognition can involve biometric data, which has stricter rules, and some uses are restricted or banned under the EU AI Act. Check with your DPO or legal advisor before any project that analyses people.
Start with ready-made tools. Custom vision models need labelled data and budget. For most SMEs, existing tools are the sensible first step.