libHIK : clustering and classifying feature vectors that are histograms
There are three components within libHIK:
A clustering method that use Histogram Intersection Kernel (HIK);
K-means and K-median clustering are also implemented in libHIK.
A method that make applying HIK SVM models extremely fast on histograms that are integers.
We also provided ways to convert real-valued histograms to integer ones. Our experiments show that there are no obvious difference between these two – the integer version works better in many cases;
Use the clustering and classification methods in the Bag of Visual words model.
Common visual codebook generation methods used in a BOV model , e.g. k-means or Gaussian Mixture Model, use the Euclidean distance to cluster features into visual code words. However, most popular visual descriptors are histograms of image measurements. It has been shown that the HIK is more effective than the Euclidean distance in supervised learning tasks with histogram features. In libHIK, we demonstrate that HIK can also be used in an unsupervised manner to significantly improve the generation of visual codebooks.
In addition, we propose a one-class SVM formulation to create more effective visual code words which can achieve even higher accuracy.
For technical details, please refer to the following paper:
Beyond the Euclidean distance: Creating effective visual codebooks\\using the histogram intersection kernel
Jianxin Wu and James M. Rehg
The twelfth IEEE International Conference on Computer Vision (ICCV 2009), Tokyo, 2009.
Version 1.5 of libHIK is available as a zip file here. Currently only C++ are supported in both Linux and Windows.
libHIK v 1.0 was published July 30 2009, updated to v 1.5 on August 10, 2009.
If you use libHIK, please cite our ICCV 2009 paper shown above. Bibtex entry of our paper is:
@InProceedings{ Wu_Rehg_2009,
title = "Beyond the Euclidean distance: Creating effective visual codebooks using the histogram intersection kernel",
booktitle = “IEEE International Conference on Computer Vision (ICCV)”,
author = "Jianxin Wu and James M. Rehg",
year = "2009"
}
Contact: Jianxin Wu (wujx2001 AT gmail.com)