MATLAB Answers


How to perform an image classification ?

Asked by Chaou
on 14 Apr 2013
Latest activity Commented on by Florian S on 6 Feb 2017

Hello. I have plenty of images, each one of them corresponds to a class. Knowing that I have 3 classes, I want to perform an image classification. I'm used to SVM and others, and I know how to perform the training and classification.

How do I proceed to the feature extraction from an image ?

Thank you!


Log in to comment.

6 Answers

Answer by Image Analyst
on 14 Apr 2013

Once you have the classified image, you essentially have a labeled image. 0 = background, a value of 1 = class 1, and so on. If you want all blobs of a certain class to be measured as a group, then just call regionprops.

groupMeasurements = regionprops(classifiedImage, 'all');

If you want each blob for a certain class to be measured by itself, then turn it into a binary image and then call regionprops:

binaryImage = classifiedImage == theClassNumberYouWant;

Then call regionprops:

individualBlobMeasurements = regionprops(binaryImage, 'all');

See my "BlobsDemo" for a more comprehensive tutorial:


Log in to comment.

Answer by Anand
on 15 Apr 2013

There are a bunch of different feature extraction approaches that are tied to what kind of images you have and the kind of problem you are attempting to solve. Here are some common feature descriptors you could extract: SURF features,MSER regions,FAST corners, Minimum Eigen Value corners and Harris corners.

Guess what, they are all in the Computer Vision System Toolbox!

Use the following functions to detect them:






and this function to extract them:


  1 Comment

Thank you for your answers! I'm dealing with this error when I run detectFASTFeatures, or any else command cited above. I put my image in the variable 'x'. He is the error:

>> size(x)

ans =

   576   720     3
>> corners = detectFASTFeatures(x); Undefined function 'detectFASTFeatures' for input arguments of type 'uint8'.

Can you help me, please?

Log in to comment.

Answer by Image Analyst
on 30 Apr 2013

Did you see the help:

"points = detectFASTFeatures(I) returns a cornerPoints object, points. The object contains information about the feature points detected in a 2-D grayscale input image, I. "

You have a 3D image - a color image. The help says that it needs to be a 2D grayscale image. Either take one color channel

grayImage = x(:,:,2);

or a weighted average of the color channels:

grayImage = rgb2gray(x);

The first method is preferable.


Log in to comment.

Answer by Chaou
on 1 May 2013
Edited by Chaou
on 1 May 2013

Hello. This is an example of 3-D images I want to work on. This image described an electrical discharge. Basically, I want to make an image recognition system able to detect near occurring discharge (as in this image), or not.

When I take one color channel, by runing this command:

grayImage = x(:,:,2);

I obtain this image:

Here are some problems, First, the 2-D image obtained does not describe the electrical discharge. Second one, even if I run this cornerPoints of the 2-D image, it still not working. It displays this message again:

*>> points = detectFASTFeatures(grayImage)* Undefined function 'detectFASTFeatures' for input arguments of type 'uint8'.

Even if I use an image from matlab library, just like this:

*>> I = imread('cameraman.tif');* *>> corners = detectFASTFeatures(I);* Undefined function 'detectFASTFeatures' for input arguments of type 'uint8'.

Can you help me, please? I would really appreciate it! :)

  1 Comment

That pseudocolored image does not make sense. Even ignoring the colormap it doesn't look anything like what it should. You should see a bright band running through that glass slide or panel or whatever it is. What does it look like if you do

imshow(grayImage, []);

Plus you need to upload an image with no discharge, and one with full discharge, and maybe one or two more with "nearly occurring discharge" so I can see what it is that distinguishes the nearly occurring discharge from the full or no discharge image.

Log in to comment.

Answer by Chaou
on 5 May 2013
Edited by Chaou
on 5 May 2013

Hello. As you requested, here are the images corresponding to no discharge, few discharges and a lot of discharges (before flash-over).



How can I extract features to detect those discharges? Thank you so much!

  1 Comment

Why don't you just get the mean gray level between certain rows? It looks like that should correlate pretty well with the level of discharge. Of course you'll want to turn off the automatic gain or exposure control in your camera because you can't have the camera trying to reduce the exposure when the image is supposed to be brighter. I can tell you have some sort of automatic gain or exposure because the intensity of the shaft on the bottom is different in the images.

Log in to comment.

Answer by Chaou
on 18 May 2013
Edited by Chaou
on 18 May 2013

Hello Image Analyst, I think that the problem is bigger. I can't even execute the Matlab tutoriel for finding corners.

>> I = imread('cameraman.tif');

>> corners = detectFASTFeatures(I);

Undefined function 'detectFASTFeatures' for input arguments of type 'uint8'.

How can I get rid of this error? Thank you so much.


I don't know. I don't have that toolbox. You might have to call them on Monday. Maybe your path is messed up. Did you do anything to your path? Can you search your hard drive in the toolbox subfolder of MATLAB for it?

I didn't touch anything on the Matlab Path. I tried those function on my friend's laptop. It displays the same error as mine. I tried to convert the image from uint8 to Double, like this:

I = imread('cameraman.tif');
I2 = im2double(I)
corners = detectHarrisFeatures(I2);
Undefined function 'detectHarrisFeatures' for input arguments of type 'double'.

Perhaps your GPU Driver is too old. I have exactly the same problems. You must have the latest CUDA driver on your running system.

I am searching for a way to perform the 'detectFASTFeatures' script without using the GPU...

Log in to comment.

Discover MakerZone

MATLAB and Simulink resources for Arduino, LEGO, and Raspberry Pi

Learn more

Discover what MATLAB® can do for your career.

Opportunities for recent engineering grads.

Apply Today

MATLAB Academy

New to MATLAB?

Learn MATLAB today!