Computer vision

The digital process of perceiving and learning visual tasks in order to interpret and understand the world through cameras and sensors.

Source: AIEOG AI Lexicon (Feb 2026), NSCAI Appendix A Technical Glossary

Computer vision is the field of AI that enables machines to interpret and process visual information from the physical world, including images, video, and real-time camera feeds. It allows AI systems to “see” and extract meaningful information from visual data.

In practice, computer vision encompasses tasks like image classification (identifying what is in a picture), object detection (locating specific items within an image), facial recognition (identifying or verifying a person’s identity), optical character recognition (reading text from images), and video analysis (understanding activity in video feeds).

Computer vision systems are typically built using deep learning techniques, specifically convolutional neural networks (CNNs) and transformer architectures trained on large datasets of labeled images.

Computer vision is used across financial services for identity verification, document processing, fraud detection, and physical security. Common applications include:

• Identity verification. Comparing a selfie to an ID document photo during customer onboarding (KYC). This is one of the most widespread financial services applications of computer vision.
• Document processing. Extracting data from physical documents: check images, loan applications, tax forms, and regulatory filings. Optical character recognition combined with computer vision reduces manual data entry.
• Check fraud detection. Analyzing check images for signs of alteration, forgery, or counterfeiting.
• ATM and branch security. Monitoring video feeds for suspicious activity, unauthorized access, or potential threats.
• Deepfake detection. Identifying AI-generated or manipulated images and videos used in fraud schemes.

Compliance considerations for computer vision are significant:

• Biometric data regulations. Facial recognition and biometric verification are subject to state and federal privacy regulations. Institutions must understand and comply with applicable biometric data laws.
• Accuracy across demographics. Computer vision systems, particularly facial recognition, have documented performance disparities across demographic groups. This creates fair lending and civil rights risk.
• Customer consent. Using computer vision for biometric purposes may require explicit customer consent depending on jurisdiction.

1. Assess demographic performance. Test computer vision systems for accuracy disparities across race, gender, age, and other demographic factors.
2. Comply with biometric data laws. Understand and comply with applicable state biometric information privacy laws (BIPA and similar).
3. Obtain appropriate consent. Ensure customer consent processes for biometric data collection meet legal requirements.
4. Validate document processing accuracy. Test OCR and document processing systems for accuracy across document types, quality levels, and languages.
5. Monitor for deepfake threats. As deepfake technology advances, ensure identity verification systems can detect AI-generated or manipulated images.
6. Include in AI governance. Computer vision systems should be included in the AI use case inventory and subject to model risk management.

Deep learning, AI model, Machine learning, Deepfake, AI system, Bias