Scene Recognition

Computer Vision | Matlab

This project uses different types of feature representation and classification to identify the scene type of a given image. A training set of 1500 labelled scenes are used as a sample set to identify the scene type of test images. The feature representation techniques used are

  • Bag of words
  • Gist along with Sift descriptors
  • Sift descriptors
  • Tiny images

And the classification techniques used are

  • Support Vector Machine
  • Naive-Bayes Nearest Neighbor
  • Nearest Neighbor classifier

The maximum accuracy achieved was 0.685.

Results visualization for good performing bag of sift – svm pipeline.


Accuracy (mean of diagonal of confusion matrix) is 0.685

Links: Project Report