Learning group-based dictionaries for discriminative image representation

Lian, Hao; Mei, Kuizhi; Zheng, Nanning; Dong, Peng; Zhou, Ning; Fan, Jianping

doi:10.1016/j.patcog.2013.07.016

Cited by 22 publications

(11 citation statements)

References 53 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…在分层分类中, 许多学者利用 ontology 知识进行分类学习, 如图像分类的 WordNet 语义结构 [24,25] 和网页分类的分层结构 [26,27] [32] 利用了近邻传播进行分层结构的聚类, 2014 年, Lei 等 [33] 采用了谱聚类构建层次结构. 2015 年, Fan 等 [34] 利用类样本的均值作为代表点度量类别之间的相似关系来建立层次结构, 虽然实现了快速的层次构建方法, 但是却丢失了较多的信息.…”

Section: 层次结构的构建unclassified

“…Griffin 等 [28] 、Feng 等 [50] 、Wen 等 [51] 、Bengio 等 [29] 、Gao 等 [18] 和 Lei 等 [33] 学者的工作. 这种自顶为了解决上述问题, 一些学者通过最小化全局损失函数的方式对层次结构中的多个分类器进行联合优化 [55∼59] .…”

Section: 给定层次结构之后分类器的学习可以通过引入层次结构信息提高性能通常来说叶子节点对应着所有的样本标签因此常unclassified

See 1 more Smart Citation

Review on hierarchical learning methods for large-scale classification task

Hu¹,

Wang²,

Zhou³

et al. 2018

Sci. Sin.-Inf.

View full text Add to dashboard Cite

Section: 层次结构的构建unclassified

Section: 给定层次结构之后分类器的学习可以通过引入层次结构信息提高性能通常来说叶子节点对应着所有的样本标签因此常unclassified

Review on hierarchical learning methods for large-scale classification task

Hu¹,

Wang²,

Zhou³

et al. 2018

Sci. Sin.-Inf.

View full text Add to dashboard Cite

“…However, hierarchical learning methods based on confusion matrices suffer from two main limitations: 1) computation of the confusion matrix using a one-vs.-all SVM is time consuming; and 2) the confusion matrix may not be reliable due to unbalanced training data. Visual trees constructed by clustering produce an intuitive hierarchical structure [12], [29], [40], [62] and have attracted more and more attention. Zhou et al [62] utilized AP clustering and Lei et al [29] implemented spectral clustering to construct visual trees.…”

Section: A Hierarchical Learningmentioning

confidence: 99%

“…Performing hierarchical inference based on the visual hierarchical structure is also challenging. Greedy learning is a typical way to solve classification prediction using visual tree models [6], [29], [62]. However, while relatively intuitive, the greedy approach does not prevent error propagation; that is, if a mistake is made in an intermediate node, the prediction result is then destined to be wrong.…”

Section: Introductionmentioning

confidence: 99%

Joint Hierarchical Category Structure Learning and Large-Scale Image Classification

Shen

et al. 2017

IEEE Trans. on Image Process.

View full text Add to dashboard Cite

We investigate the scalable image classification problem with a large number of categories. Hierarchical visual data structures are helpful for improving the efficiency and performance of large-scale multi-class classification. We propose a novel image classification method based on learning hierarchical inter-class structures. Specifically, we first design a fast algorithm to compute the similarity metric between categories, based on which a visual tree is constructed by hierarchical spectral clustering. Using the learned visual tree, a test sample label is efficiently predicted by searching for the best path over the entire tree. The proposed method is extensively evaluated on the ILSVRC2010 and Caltech 256 benchmark datasets. The experimental results show that our method obtains significantly better category hierarchies than other state-of-the-art visual tree-based methods and, therefore, much more accurate classification.

show abstract

“…The BOVW is used to represent local features and descriptors, along with geometry verification, which is motivated by an analogy, with the 'bag-of-words' representation for text categorization. There are publications [4]- [7] about visual content representation using the BOVW due to it being a promising method for visual content classification, annotation, and retrieval. The BOVW model of images may be classified in a class on the basis of visual word histograms.…”

Section: Introductionmentioning

confidence: 99%

The Selection of Useful Visual Words for Class-Imbalanced Data in Image Classification

Chimlek

Pramokchon

Piamsa-nga

2016

IJECE

View full text Add to dashboard Cite

<span>The bag of visual words (BOVW) has recently been used for image classification in large datasets. A major problem of image classification using BOVW is high dimensionality, with most features usually being irrelevant and different BOVW for multi-view images in each class. Therefore, the selection of significant visual words for multi-view images in each class is an essential method to reduce the size of BOVW while retaining the high performance of image classification. Many feature scores for ranking produce low classification performance for class imbalanced distributions and multi-views in each class. We propose a feature score based on the statistical t-test technique, which is a statistical evaluation of the difference between two sample means, to assess the discriminating power of each individual feature. The multi-class image classification performance of the proposed feature score is compared with four modern feature scores, such as Document Frequency (DF), Mutual information (MI), Pointwise Mutual information (PMI) and Chi-square statistics (CHI). The results show that the average F1-measure performance on the Paris dataset and the SUN397 dataset using the proposed feature score are 92% and 94%, respectively, while all other feature scores do not exceed 80%.</span>

show abstract

Learning group-based dictionaries for discriminative image representation

Cited by 22 publications

References 53 publications

Review on hierarchical learning methods for large-scale classification task

Review on hierarchical learning methods for large-scale classification task

Joint Hierarchical Category Structure Learning and Large-Scale Image Classification

The Selection of Useful Visual Words for Class-Imbalanced Data in Image Classification

Contact Info

Product

Resources

About