模式识别 - 处理多示例学习(MIL)特征(matlab)

处理多示例学习(MIL)特征(matlab)

本文地址: http://blog.csdn.net/caroline_wendy/article/details/27206325

多示例学习(MIL)的特征(features)包含, "示例编号+视频编号+标签+特征"的形式;

需要组成多示例学习特定包的形式, 每一个元胞是一个多示例包, 后面一位是标签, 即"特征包+标签"的形式;

代码:

%author @ C.L.Wang
%time @ 2014.5.27
%matlab @ R2012a

%% 加载数据
clc, clear;
features = load('ViolenceMILFeatures.txt');

%% 提取多示例包
head = features(:,1:3); % 头部信息
video_vector = features(:,2); % 视频名称
total_num = size(unique(video_vector),1);
video_cell = cell(total_num,2);
for video_num = 1:total_num
    temp = find(video_vector==video_num);
    video_cell{video_num,1} = features(temp(1,1):temp(end,1),4:end);
    video_cell{video_num,2} = features(temp(1,1),3);
end

%% 分类训练和测试
negative_train_features = video_cell(1:(total_num/4), :);
negative_test_features = video_cell((total_num/4+1):(total_num*2/4), :);
positive_train_features = video_cell((total_num*2/4+1):(total_num*3/4), :);
positive_test_features = video_cell((total_num*3/4+1):end, :);

train_features = [positive_train_features; negative_train_features]; %整合训练
test_features = [positive_test_features; negative_test_features]; %整合测试