Last reviewed:
What is computer vision? Definition, examples and the AI Act framework
Computer vision is the field of artificial intelligence that enables a machine to analyse images and video: recognising objects, detecting defects, reading a document, tracking movement. In Europe, its biometric uses (facial recognition, emotion recognition) are partly prohibited and partly classified as high-risk by the AI Act.
Computer vision turns pixels into usable information. Its basic tasks are classification (what does the image show?), detection (where is each object?), segmentation (which pixels belong to which object?) and optical character recognition, or OCR (what text appears on the document?). The turning point came in 2012: the AlexNet convolutional neural network, presented at the NIPS conference, classified 1.3 million images into 1,000 categories with an error rate well below the state of the art of the time, launching the rise of deep learning in vision. Since then, multimodal models (Claude, GPT, Gemini) also read images: they describe a photo, extract data from an invoice or read a screenshot from a simple instruction, where a dedicated model trained on thousands of annotated images used to be needed. Business applications are numerous: quality control on a production line, automatic reading of invoices and ID documents, in-store inventory, safety on construction sites, waste sorting. The AI Act mainly regulates biometric uses. Since February 2, 2025, Article 5 prohibits in particular building facial recognition databases through untargeted scraping of images from the internet or CCTV, emotion recognition in the workplace and in education (except for medical or safety reasons) and, with strict exceptions, real-time remote biometric identification in publicly accessible spaces for law enforcement. Remote biometric identification and emotion recognition are also high-risk (Annex III).
Concrete example
Illustrative case: a French packaging manufacturer with 250 employees installs cameras at the end of the line. A model trained on 20,000 photos of defects (burrs, misalignments, scratches) rejects non-compliant parts in real time. Quality inspectors now only check doubtful cases flagged by the system, and the data collected helps identify the machine that is drifting. The project processes no biometric data: the cameras point at the products, not at the operators.
Real case: for the Paris 2024 Olympic Games, Article 10 of the French law of May 19, 2023 authorised, on an experimental basis, algorithmic analysis of video surveillance images to detect risk events at large gatherings. The text excludes any biometric identification, any biometric data and any facial recognition. A law of August 18, 2026 extended the experiment until December 31, 2030.
Comparison
| Task | What the model does | Business example | AI Act framework |
|---|---|---|---|
| Classification | Assigns a category to the image | Compliant or defective part | Generally no specific obligation |
| Object detection | Locates each object in the image | Counting products on shelves, hard hat detection | Generally no specific obligation |
| OCR and document reading | Extracts text and fields | Automatic invoice data entry | Generally no specific obligation |
| Remote biometric identification | Recognises a person by face or silhouette | Remote camera-based access control | High-risk (Annex III); prohibited in real time in public spaces for law enforcement, with exceptions |
| Emotion recognition | Infers an emotional state | Analysing customer reactions | High-risk; prohibited at work and in education |
FAQ
What is computer vision?
It is the branch of AI that enables a machine to analyse images and video: recognising objects, detecting anomalies, reading text, tracking movement. It relies mainly on neural networks trained on large quantities of annotated images.
What are examples of computer vision in business?
Quality control on a production line, automatic reading of invoices and delivery notes, ID document verification, shelf and stock monitoring in stores, detection of safety equipment on construction sites, automated waste sorting, support for medical imaging diagnosis.
What is the difference between computer vision and image processing?
Image processing modifies or improves an image (sharpness, contrast, compression). Computer vision extracts an interpretation from it: what the image contains, where, and what it means. Image processing is often a preliminary step.
Is facial recognition allowed in Europe?
It is tightly regulated. Since February 2025, the AI Act prohibits building face databases through untargeted scraping and, with strict exceptions, real-time biometric identification in public spaces for law enforcement. Other remote biometric identification systems are classified as high-risk. The GDPR also protects biometric data.
How does computer vision relate to multimodal AI?
Multimodal models integrate computer vision into a language model: you can show them an image and ask a question in plain language. For a massive, very precise task, such as quality control at industrial speed, a dedicated vision model is often faster and cheaper.
Where can I learn computer vision (courses)?
For an executive, the key is to understand the tasks (classification, detection, segmentation, OCR) and their limits. For a technical team, online courses from universities and vendors, along with open libraries such as OpenCV, are the usual entry points.
See also
Further reading
Article 5 of the AI Act (prohibited AI practices), AI Act Service Desk, European Commission
Sources
- ImageNet Classification with Deep Convolutional Neural Networks, Krizhevsky, Sutskever and Hinton, NeurIPS (NIPS) 2012. https://proceedings.neurips.cc/paper_files/paper/2012/hash/c399862d3b9d6b76c8436e924a68c45b-Abstract.html
- Regulation (EU) 2024/1689 on artificial intelligence, Article 5 (prohibited practices), AI Act Service Desk, European Commission. https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-5
- Regulation (EU) 2024/1689 on artificial intelligence, Annex III, point 1 (biometrics), AI Act Service Desk, European Commission. https://ai-act-service-desk.ec.europa.eu/en/ai-act/annex-3
- French law no. 2023-380 of May 19, 2023 on the 2024 Olympic and Paralympic Games, Article 10, as amended by law no. 2026-798 of August 18, 2026, Légifrance. https://www.legifrance.gouv.fr/loda/article_lc/LEGIARTI000047564439