readme.txt

论文Beyond Bags of Features源代码

========================================================================
Spatial Pyramid Code
Created by Joe Tighe ([e-mail]) and Svetlana Lazebnik ([e-mail])
1/17/2009

This MATLAB code implements spatial pyramid computation and matching as 
described in the following paper:

S. Lazebnik, C. Schmid, and J. Ponce, ``Beyond Bags of Features: Spatial 
Pyramid Matching for Recognizing Natural Scene Categories,'' CVPR 2006.

========================================================================


The main function to build the spatial pyramid is BuildPyramid.
For further information on Buildpyramid and other functions discussed in 
this file refer to comments in the .m files or look at Example.m.
(The images/ directory contains a few sample images that are used by 
Example.m to compute spatial pyramids and their histogram intersection
matrix.)

BuildPyramid first extracts SIFT descriptors on a regular grid from each 
image. It then runs k-means to find the dictionary. Each sift descriptor 
is given a texton label corresponding to the closest dictionary codeword. 
Finally, the spatial pyramid is generated from these labels.

Each of these steps are split up into individual functions and can be called 
independently, provided the data from the previous step is stored in the 
correct location. The functions are as follows:

GenerateSiftDescriptors
CalculateDictionary
BuildHistograms
CompilePyramid

If you wish to use one of these functions without first running the previous 
step, you will need to provide the appropriate data files. They should be in 
the dataBaseDir with the same relative path as the image they correspond to. 
Their file names should be the same as the image they correspond to with the 
appropriate suffix appended to the end. For instance if you call 
CalculateDictionary with featureSuffix = 慱sift.mat