Bag of Word主要思想:将训练样本特征Kmeans聚类,对测试样本的每个特征,计算与其最近的类心,相应类别计数count加1,这样每个测试样本可以生成ncenter维的直方图。
比如:训练样本特征a、b、c、a、d、f、e、b、e、d、c、f,如果类别数ncenter为6,则可以聚成6类[a,b,c,d,e,f]注意实际聚类时类心不一定为训练样本中特征,因为kmeans聚类更新类心时都重新计算。
假如一个测试样本特征为:a、b、c、d.那么经过BoW生成6维的直方图[1,1,1,1,0,0].
其实前面就是kmeans,然后Hard voting。关于kmeans不细说了,就是更新类心的过程,一直到类心变化在误差范围内。
kmeans聚类时用的训练数据中center个随机数据初始化,后面用的欧氏距离度量,其中计算欧氏距离时用了矢量化编程,加速运算。
这是参考了别人的代码实现的,每个人针对自己的研究可能还需要小小修改。适合入门的看看。
function dic=CalDic(data,dicsize)
fprintf('Building Dictionary using Training Data\n\n');
dictionarySize = dicsize;
niters=100;%迭代次数
centres=zeros(dictionarySize,size(data,2));
[ndata,data_dim]=size(data);
[ncentres,dim]=size(centres);
%% initialization
perm = randperm(ndata);
perm = perm(1:ncentres);
centres = data(perm, :);
num_points=zeros(1,dictionarySize);
old_centres = centres;
display('Run k-means');
for n=1:niters
% Save old centres to check for termination
e2=max(max(abs(centres - old_centres)));
inError(n)=e2;
old_centres = centres;
tempc = zeros(ncentres, dim);
num_points=zeros(1,ncentres);
[ndata, data_dim] = size(data);
id = eye(ncentres);
d2 = EuclideanDistance(data,centres);
% Assign each point to nearest centre
[minvals, index] = min(d2', [], 1);
post = id(index,:); % matrix, if word i is in cluster j, post(i,j)=1, else 0;
num_points = num_points + sum(post, 1);
for j = 1:ncentres
tempc(j,:) = tempc(j,:)+sum(data(find(post(:,j)),:), 1);
end
for j = 1:ncentres
if num_points(j)>0
centres(j,:) = tempc(j,:)/num_points(j);
end
end
if n > 1
% Test for termination
%Threshold
ThrError=0.009;
if max(max(abs(centres - old_centres))) <0.009
dictionary= centres;
fprintf('Saving texton dictionary\n');
mkdir('data');%建立data文件夹
save ('data\dictionary','dictionary');%保存dictionary到data文件夹下。
break;
end
fprintf('The %d th interation finished \n',n);
end
end
下面是欧氏距离函数:
function d = EuclideanDistance(a,b)
% DISTANCE - computes Euclidean distance matrix
%
% E = EuclideanDistance(A,B)
%
% A - (MxD) matrix
% B - (NxD) matrix
%
% Returns:
% E - (MxN) Euclidean distances between vectors in A and B
%
%
% Description :
% This fully vectorized (VERY FAST!) m-file computes the
% Euclidean distance between two vectors by:
%
% ||A-B|| = sqrt ( ||A||^2 + ||B||^2 - 2*A.B )
%
% Example :
% A = rand(100,400); B = rand(200,400);
% d = EuclideanDistance(A,B);
% Author : Roland Bunschoten
% University of Amsterdam
% Intelligent Autonomous Systems (IAS) group
% Kruislaan 403 1098 SJ Amsterdam
% tel.(+31)20-5257524
% bunschot@wins.uva.nl
% Last Rev : Oct 29 16:35:48 MET DST 1999
% Tested : PC Matlab v5.2 and Solaris Matlab v5.3
% Thanx : Nikos Vlassis
% Copyright notice: You are free to modify, extend and distribute
% this code granted that the author of the original code is
% mentioned as the original author of the code.
if (nargin ~= 2)
b=a;
end
if (size(a,2) ~= size(b,2))
error('A and B should be of same dimensionality');
end
aa=sum(a.*a,2); bb=sum(b.*b,2); ab=a*b';
d = sqrt(abs(repmat(aa,[1 size(bb,1)]) + repmat(bb',[size(aa,1) 1]) - 2*ab));
下面是Hard Voting函数:
function His=HardVoting(data,dic)
ncentres=size(dic,1);
id = eye(ncentres);
d2 = EuclideanDistance(data,dic);% Assign each point to nearest centre
[minvals, index] = min(d2', [], 1);
post = id(index,:); % matrix, if word i is in cluster j, post(i,j)=1, else 0
His=sum(post, 1);
end
如果用于分类问题,可以尝试用LLC(CVPR2010) 一般比Hard Voting效果好。