GistEstimatorContextBased.H File Reference

Implementation of ``Context-based vision system for place and object recognition'' by Torralba, et al. More...

#include "Neuro/GistEstimator.H"
#include "Neuro/NeuroSimEvents.H"
#include <string>

Include dependency graph for GistEstimatorContextBased.H:

This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Classes
class	GistEstimatorContextBased
	Gist estimator for ``Context-based vision system...'' by Torralba, et al. More...

---

Detailed Description

Implementation of ``Context-based vision system for place and object recognition'' by Torralba, et al.

The GistEstimatorContextBased class implements (within the INVT framework) the gist related portions of the following paper:

Torralba, A., Murphy, K. P., Freeman, W. T., Rubin, M. A. Context-based vision system for place and object recognition. Ninth IEEE International Conference on Computer Vision, 1:273--280, 2003.

In the paper, the authors use a wavelet image decomposition tuned to 6 orientations and 4 scales to compute texture features. However, this class uses Gabor filters (with the same number of orientations and scales) because INVT readily provides them and, as per the authors, the two approaches produce similar results.

To compute the gist vector for an image, we apply the 24 Gabor filters and then subdivide each resulting image into a 4x4 grid. Then, we compute the mean pixel value in each grid. This gives us 16 numbers per filter. Since there are 24 filters in all, we get 16x24 = 384 numbers per image. These 384 numbers form the gist vector for the input image.

Definition in file GistEstimatorContextBased.H.