Computer Vision and Image Generation
Level Up Team
This post summarizes Computer Vision and Image Generation, module 2.4 of Level Up’s Azure AI Fundamentals series. The lesson is on YouTube as Azure AI Fundamentals AI-901 Exam: 2.4: Computer Vision and Image Generation. Module 2.3 is Speech AI.
Start with the output, not the image
A product photograph can support classification, detection, optical character recognition, interpretation, or generation. The required output determines the capability. Possible outputs are a label, located objects, visible text, an interpretation, or a new image.
Image classification labels the whole image
Image classification assigns one or more labels to an image. A product photograph might be labeled “damaged laptop.” Teams use it to identify a species, categorize a product, detect whether an image contains damage, or classify manufacturing defects.
Classification answers “What is this image?” It does not return object locations, and it does not generate a new picture. The exam signal is which category best describes the image.
Object detection finds items and where they are
Object detection identifies objects and returns their locations. A typical result includes an object label, a confidence score, and bounding-box coordinates: a laptop at one region of the frame, and a crack at another.
Classification answers what the image is. Object detection finds multiple objects and where they appear. The exam signal is identifying where each object appears.
OCR reads the characters that are visible
Optical character recognition converts text visible in an image or scanned document into machine-readable text. A photograph of a shipping label becomes “Order 10482.”
Common uses are signs and labels, scanned pages, serial numbers, and printed or handwritten text. OCR tells you which characters are present. It does not necessarily explain the document or map the text into business fields. The exam signal is reading or digitizing the text shown in an image.
Image understanding interprets the scene
Image understanding uses visual context to describe an image, explain what is happening, compare visual elements, answer questions about the scene, or reason across images and text.
An interpretation might be: “The laptop screen appears cracked near the upper-right corner, but the shipping box shows no visible exterior damage.” That is different from reading characters or returning invoice fields. The exam signal is describe, interpret, compare, or answer a question about visual content.
Image generation creates a new picture
Image generation creates a new image from a prompt or other input. “Create a clean product-support illustration of a laptop with a cracked screen” produces a new image, an edit, an extension, or a variation. Teams use it for marketing artwork, product concepts, illustrations, and edits of an existing image.
If the scenario analyzes an existing image, the workload is computer vision. If it produces a new image, the workload is image generation.
The same photograph can demand four different capabilities
A photograph of an invoice is not automatically one workload.
| Desired output | Capability |
|---|---|
| Read all visible words | OCR |
| Describe the document | Image understanding |
| Determine whether it is an invoice | Image classification |
| Return vendor, date, and total as fields | Information extraction |
Information extraction can use OCR and vision internally. The business goal is different: defined fields or records, such as vendor “Contoso,” invoice date “2026-09-12,” and total 847.50. Understanding explains the image. Extraction maps its content into a schema.
Classify the output
| Required output | Capability |
|---|---|
| A category | Image classification |
| Objects plus locations | Object detection |
| Visible characters | OCR |
| A description or an answer | Image understanding |
| Structured fields | Information extraction |
| A new visual | Image generation |
A claims system that receives photographs of vehicle damage and must identify each damaged component and mark its position needs object detection. Classification would label the image. It would not show where each damaged component appears.
Bring this to your team
Level Up can run a personalized Azure AI Fundamentals workshop for your team. Tell us a little about your group and we’ll reply with pricing.