基于多种输出嵌入结合的无标签图像分类

利用多种输出嵌入相结合的方法,改善无标签图像的分类性能.以边信息作为标签嵌入,用图像特征作为输入嵌入,在标签嵌入和输入嵌入之间构建一个联合兼容函数,建立结构化联合嵌入框架.通过调整联合嵌入的权重矩阵,使兼容函数取得最大值,据此确定图像的分类.借助两个数据集进行的验证,实验结果显示,多种输出嵌入结合的图像分类方法准确率优于单输出嵌入的图像分类方法.

Unlabeled image classification based on multiple output embeddings

By using the method of combining multiple output embeddings, the performance of unlabeled image classification is improved. Side information is used as label embedding and image features are used as output embedding, by introducing a joint compatibility function between label embedding and output embedding, the Structured Joint Embedding framework is established. By adjusting the weighting matrix to make the compatibility function to the maximum, and thus the image classification is determined. Validation with two data sets, the experiment results show that the image classification method of combining multiple output embeddings has superior accuracy to that of using the single output embedding.