0
$\begingroup$

In my final year project, we are using a SIFT feature descriptor to classify and recognize objects, but I have never come across segmentation in the process of image classification/categorization.

Isn't this step necessary?

$\endgroup$
  • $\begingroup$ Hey, if I understand what you're asking, then, no, segmentation is not a necessary step in the process of classification, categorization or object detection (although it can sometimes be used). If you want some more information, please specify which part of the task pipeline is confusing for you (and possibly the way you understood it) and we'll try to help. If you're asking something else then your question is somewhat unclear... $\endgroup$ – penelope Nov 18 '13 at 9:34
1
$\begingroup$

SIFT works on points in an image while segmentation is about dividing up the image into regions. So, no, segmentation is not necessary when using SIFT.

In segmentation you divide up the image into regions so classification can be done by extracting features from each region and see if you can recognize your object. A downside to this approach is that producing a good segmentation is a difficult problem. Many segmentation techniques are sensitive to various circumstances, such as lighting, contrast, colors etc. These difficult situations often result in regions that do not represent your objects very well.

In SIFT you focus on specific points (keypoints) in the image instead of entire regions. If you want to recognize an object in an image using SIFT, you extract keypoints from an image of your object and give each keypoint a description (keypoint descriptor) based on its local neighborhood. Then you check if similar keypoints exist in the image you want to analyze. By matching the keypoints between your object image and the image you want to analyze you can do object recognition without segmentation.

$\endgroup$
  • $\begingroup$ Thanks a lot mags.One more doubt,you said that we use SIFT to extract keypoints from an image,here in the paper published by Lowe,a DoG(Difference of Gaussian) method is suggested to get these points.I understood that DoG detects changes in the intensities (corner,blob or edge) but what I can't visualize is the fact that how can it be used to describe certain features present in that image. Thanks and regards in Advance $\endgroup$ – logamadi Nov 19 '13 at 17:48
  • $\begingroup$ DoG is used to find the SIFT keypoints. Locating the keypoints is considered a part of the SIFT. Lots of different strategies can be used to locate the keypoints and DoG is a popular method for doing this. Once the key points have been found, other strategies are used to create the description of each point. $\endgroup$ – mags Nov 19 '13 at 20:30
  • $\begingroup$ @mags.I was wondering whether this detector is exhaustive to determine all the required features in the image.How can we decide whether this is the best detector. $\endgroup$ – logamadi Nov 20 '13 at 4:10

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.