Efficient Feature Coding Based on Auto-encoder Network for Image Classification
Efficient Feature Coding Based on Auto-encoder Network for Image Classification
复制标题
DOI：
10.1007/978-3-319-16865-4_41
复制
发表时间：
2014-11
期刊：
影响因子：
--
通讯作者：
Guosen Xie;Xu-Yao Zhang;Cheng-Lin Liu
中科院分区：
文献类型：
--
作者：
Guosen Xie;Xu-Yao Zhang;Cheng-Lin Liu
局部描述符编码是传统词袋(BOW)图像分类框架中的关键一步。然而，以往方法的编码速度慢是限制大规模问题应用的一个因素。近年来，神经网络模型在各种分类任务中得到了广泛的应用。由于其快速前向传播，使用神经网络模型进行描述符编码是简单而有效的。本文提出将自动编码器(AE)网络作为局部描述符编码块，并将其嵌入到BOW框架中以达到图像分类的目的。为了使AE网络的隐含活动同时具有选择性和稀疏性，我们在AE网络的学习过程中加入了一个高效有效的正则化项，从而提高了每个输入描述子隐含层的稀疏性和每个隐含节点的选择性。通过将AE网络编码与BOW框架相结合，我们可以在Caltech101、Scene15和UIUC 8-Sports数据库上获得比其他最先进的功能编码方法更好的结果和更快的速度。
Local descriptor coding is one crucial step in traditional Bag of Words (BoW) framework for image categorization. However, the slow coding speed of previous methods is one limitation for applications in large scale problems. Recently, neural network based models have been widely applied in various classification tasks. Using neural network models for descriptor coding is straightforward and efficient due to their fast forward propagation. In this paper, we propose to use the Auto-Encoder (AE) network as a local descriptor coding block, and further embed AE network in the BoW framework for the purpose of image classification. To make the hidden activities of AE network to be both selective and sparse, we add an efficient and effective regularization term into the learning process of AE network, which can promote sparsity of the hidden layer for each input descriptor as well as the selectivity for each hidden node. By incorporating the AE network coding with the BoW framework, we can achieve better results and faster speeds than other state-of-the-art feature coding methods on Caltech101, Scene15 and UIUC 8-Sports databases.