Ask Your Question

image segmentation

asked 2020-05-28 08:15:40 -0500

dineshlama gravatar image

According to what i've found till now, the difinition of image segmentation is described as

In digital image processing and computer vision, image segmentation is the process of partitioning a digital image into multiple segments

Here it says, partifitoning or say deviding a digital image into multiple segments. As i saw it is basically done based on different colour of different regions. But what is the exact goal of image segmentation and how can it be defined in sentence? Eg. If it is based on colour than even a same region of object eg. apple can have different colour values in its surface due to lots of reasons like lightening condition etc. So can someone explain me the exact definition what is the purpose of image segmentation i.e exact output i.e expected by it?

edit retag flag offensive close merge delete



As i saw it is basically done based on different colour of different regions.

no, that's only the most silly way to try so. more successful algorithms use cnns / gradient based graphs and whatnot.

berak gravatar imageberak ( 2020-05-28 09:38:34 -0500 )edit

1 answer

Sort by ยป oldest newest most voted

answered 2020-05-28 10:18:20 -0500

kbarni gravatar image

updated 2020-05-28 10:30:14 -0500

As the definition above says, segmentation is partitioning the image into different regions.

You can imagine these regions as parts of the image representing for example different objects.

This can be exhaustive (all the objects are identified), for example when analyzing the environment of self driving cars: the road, sidewalk, traffic signs, pedestrians, cars, bicycles, trees, buildings etc. are identified. Here's an example from nvidia dev blog: image description

It can also be an object against the background (or anything else). For example when trying to identify a cancer on a radiography image. Another image from NVIDIA: image description

The segmented regions don't have to represent objects. You can segment colors (example application: how ripe (red) is an apple?), textures (e.g. cracks/defects in a material, seismic image processing), or anything else (e.g. detect the parts of a photo manipulated by Photoshop).

As you can see, in the simple situations the segmentation can be color-based (see thresholding) or gradient/edge-based (e.g.canny edge detector), you can rely on textures, and in complicated situations you'll need to use deep neural networks (segnets) and so on...

edit flag offensive delete link more


The definition is a good one. How you segment / on what criteria is based on what you want to. It can be objects / colors / shapes and many other things.

holger gravatar imageholger ( 2020-05-28 13:51:54 -0500 )edit
Login/Signup to Answer

Question Tools

1 follower


Asked: 2020-05-28 08:15:40 -0500

Seen: 353 times

Last updated: May 28