0

I am using SVM Light to classify images that are processed with OpenCV. Images are made to B/W, blurred a little, and the HOG detector from opencv is used to create a feature vector with vectors from positive images labeled with a 1 and negative images with a -1. When I run the SVMLight train file on 7 positive and 7 negative process images, it misclassifies 4 of the 7 negative files.

However, at larger inputs, it trains without misclassifications. Does anyone know why this may be the case?

Correctly classifying negative images with a larger training input

Misclassifying 4 of 7 negative images while training

Stephanie Mertz
  • 1
  • 1

1 Answers1

0

This is one of the disadvantages of SVMs in general. For instance in the scikit-learn documentation, it's mentioned:

The disadvantages of support vector machines include:

  • If the number of features is much greater than the number of samples, the method is likely to give poor performances.
Community
  • 1
  • 1
fireant
  • 14,080
  • 4
  • 39
  • 48