BagOfWords

Bag of words class

The BagOfWords class holds sets of features for a number of images and supports image retrieval by comparing new images with those in the 'bag'.

Methods

Properties

Reference

J.Sivic and A.Zisserman, "Video Google: a text retrieval approach to object matching in videos", in Proc. Ninth IEEE Int. Conf. on Computer Vision, pp.1470-1477, Oct. 2003.

See also

PointFeature

Create a BagOfWords object

b = BagOfWords(f, k) is a new bag of words created from the feature vector f and with k words. f can also be a cell array, as produced by ISURF() for an image sequence.

The features are sorted into k clusters and each cluster is termed a visual word.

b = BagOfWords(f, b2) is a new bag of words created from the feature vector f but clustered to the words (and stop words) from the existing bag b2.

Notes

• Uses the MEX function vl_kmeans to perform clustering (vlfeat.org).

See also

PointFeature, isurf

Convert to string

s = B.char() is a compact string representation of a bag of words.

Find images containing word

k = B.contains(w) is a vector of the indices of images in the sequence that contain one or more instances of the word w.

Display value

B.display() displays the parameters of the bag in a compact human readable form.

Notes

• This method is invoked implicitly at the command line when the result of an expression is a BagOfWords object and the command has no trailing semicolon.

See also

BagOfWords.char

Display exemplars of words

B.exemplars(w, images, options) displays examples of the support regions of the words specified by the vector w. The examples are displayed as a table of thumbnail images. The original sequence of images from which the features were extracted must be provided as images.

Options

Features from words

f = B.isword(w) is a vector of feature objects that are assigned to any of the word w. If w is a vector of words the result is a vector of features assigned to all the words in w.

Word occurrence

n = B.occurrence(w) is the number of occurrences of the word w across all features in the bag.

Remove stop words

B.remove_stop(n) removes the n most frequent words (the stop words) from the bag. All remaining words are renumbered so that the word labels are consecutive.

Word frequency statistics

[w,n] = B.wordfreq() is a vector of word labels w and the corresponding elements of n are the number of occurrences of that word.

Word frequency vector

wf = B.wordvector(J) is the word frequency vector for the J'th image in the bag. The vector is Kx1 and the angle between any two WFVs is an indication of image similarity.

Notes

• The word vector is expensive to compute so a lazy evaluation is performed on the first call to this function

Machine Vision Toolbox for MATLAB © 1990-2014 Peter Corke.