What is image segmentation in deep learning?

Image segmentation is a deep learning task that assigns a class or label to individual pixels in an image.

Instead of only identifying what is present in the image, segmentation also shows the exact region occupied by an object or category.

How image segmentation works

A segmentation model takes an image as input and produces a pixel-level output, often called a segmentation mask.

For example:

Input image → Segmentation model → Pixel-level mask

Pixels belonging to the same object or class are grouped together in the output.

Main types of image segmentation

• Semantic segmentation: assigns every pixel to a class, such as road, car, sky, or person.
• Instance segmentation: separates different objects even when they belong to the same class.

For example, if an image contains three people:

• Semantic segmentation labels all three as person.
• Instance segmentation distinguishes person 1, person 2, and person 3 separately.

How is segmentation different from classification and object detection?

• Image classification: predicts a label for the whole image.
• Object detection: predicts object classes and bounding boxes.
• Image segmentation: predicts object or class regions at the pixel level.

This makes segmentation useful when precise object boundaries matter.

Where is image segmentation used?

Common applications include:

• Medical imaging
• Autonomous driving
• Satellite and aerial imagery
• Robotics
• Industrial inspection
• Background or foreground separation

For related tasks, see “What is image classification in deep learning?” and “What is object detection in deep learning?”.

In short, image segmentation labels pixels so a model can identify not only what appears in an image, but the exact regions where objects or classes are located.