Tips on global features in images that are usable in SVM

Hello,

I am currently working on a project where i am classifying images using a SVM. Since I cant reach a high enough result with only HOG or Gaborwavelet so i was thinking i was going to add a third one and then make a decisionmaking program where the majority will decide which class the image belongs to.

So the the question; what global feature do you recommend? They basically HAVE to be global, otherwise the featurevectors dimensions missmatch witch will mess up the result.

Thanks in advance, Anton

[Image Analyst]

Why don't you attach a few images and tell us what class(es) they belong to? As you noticed, it's key to identify the proper features to classify images - you can't use just any old random measurement.

Answer 1

[Nalini Vishnoi]

Anton,

You can use color (histograms) and/or texture based features in addition to HOG and Gaborwavelet. These would be global in nature. However, it is heavily dependent on your data set what type of features would be most distinguishable from each other.

Nalini