Overview

Image classification is one of the most common computer vision workloads assessed on the AI-900 exam. It focuses on assigning one or more labels to an image based on its visual content. Unlike object detection, image classification does not identify locations within the image — it answers the question:

“What is this image?”

On the AI-900 exam, you are expected to recognize when image classification is the correct solution, understand its core features, and know which Azure services support it.

---

What Is Image Classification?

Image classification is a computer vision technique that analyzes an image and categorizes it into predefined classes or labels.

Key Characteristics

• Operates on the entire image
• Produces labels or categories
• Does not draw bounding boxes
• Often uses deep learning models (convolutional neural networks)

Simple Examples

• Classifying photos as cat, dog, or bird
• Determining whether an image contains food, landscape, or people
• Categorizing medical images as normal or abnormal

---

Common Image Classification Scenarios

Image classification is appropriate when the goal is overall categorization, not detailed localization.

Typical Use Cases

• Product categorization (e.g., retail images)
• Content moderation (safe vs unsafe images)
• Quality inspection (defective vs non-defective)
• Medical imaging classification
• Scene recognition (indoor vs outdoor)

---

Image Classification vs Other Computer Vision Tasks

Understanding how image classification differs from related workloads is critical for the AI-900 exam.

Task	What It Does
Image classification	Assigns labels to an entire image
Object detection	Identifies and locates objects with bounding boxes
Image segmentation	Classifies each pixel in an image
Facial recognition	Identifies or verifies people

Exam Tip:
If the question mentions counting, locating, or drawing boxes, image classification is not the correct answer.

---

Azure Services for Image Classification

On the AI-900 exam, Microsoft primarily expects familiarity with Azure AI Vision and Custom Vision.

Azure AI Vision (Prebuilt Models)

• Provides ready-to-use image classification
• Can identify:
  • Objects
  • Scenes
  • Tags
• Requires no model training
• Ideal for general-purpose scenarios

Azure AI Custom Vision

• Allows you to train your own image classification model
• Supports:
  • Custom labels
  • Domain-specific images
• Requires labeled training data
• Useful when prebuilt models are insufficient

---

Features of Image Classification Solutions

1. Label-Based Output

Image classification solutions return:

• One or more labels
• Confidence scores for each label

Example output:

• Dog – 92%
• Animal – 99%

---

2. Whole-Image Analysis

• The model evaluates the entire image
• No spatial location information is returned

This is a common AI-900 trick — don’t confuse classification with detection.

---

3. Confidence Scores

Predictions are typically accompanied by:

• Probability or confidence values
• Useful for decision-making thresholds

---

4. Model Training Options

Depending on the service:

• Prebuilt models require no training
• Custom Vision models require:
  • Labeled images
  • Training and evaluation cycles

---

5. Cloud-Based Inference

Azure image classification solutions:

• Run in the cloud
• Are accessed via REST APIs
• Scale automatically

---

When to Use Image Classification

Image classification is the best choice when:

• You only need to know what is in the image
• Object location is not required
• Labels are predefined or can be trained

When Not to Use It

• When you need to count objects
• When you need bounding boxes
• When identifying specific individuals

---

Responsible AI Considerations

While AI-900 does not go deep technically, you should understand high-level considerations:

• Bias in training images can affect predictions
• Transparency in how labels are applied
• Privacy concerns when images contain people

---

Key Exam Takeaways

• Image classification assigns labels to entire images
• It does not locate or count objects
• Azure AI Vision and Custom Vision are the primary services
• Look for keywords like categorize, classify, label
• Be careful not to confuse classification with object detection

---

Go to the Practice Exam Questions for this topic.

Go to the AI-900 Exam Prep Hub main page.