新科学想法 学术文库 学术文献 浏览文献

有读书笔记有附件Learning Discriminative and Shareable Features for Scene Classification

mengyang1983 添加于 2014-12-21 10:17 | 2156 次阅读 | 0 个评论
  •  作 者

    Zhen Zuo, Gang Wang, Bing Shuai, Lifan Zhao, Qingxiong Yang, and Xudong Jiang
  •  摘 要

    In this paper, we propose to learn a discriminative and shareable feature transformation filter bank to transform local image patches (represented as raw pixel values) into features for scene image classification. The learned filters are expected to: (1) encode common visual patterns of a flexible number of categories; (2) encode discriminative and class-specific information. For each category, a subset of the filters are activated in a data-adaptive manner, meanwhile sharing of filters among different categories is also allowed. Discriminative power of the filter bank is further enhanced by enforcing the features from the same category to be close to each other in the feature space, while features from different categories to be far away from each other. The experimental results on three challenging scene image classification datasets indicate that our features can achieve very promising performance. Furthermore, our features also show great complementary effect to the state-of-the-art ConvNets feature.
  •  详细资料

    • 关键词: cs.CV
    • 文献种类: Conference
    • 会议: ECCV
    • 期卷页: 2014
    • 日期: 2014-2-24
    • 发布方式: arXiv e-prints
    • 备注:arXiv:1312.6229v4
  • 学科领域 信息系统 » 计算机科学

  • 相关链接  URL 

  •  mengyang1983 的文献笔记  订阅

    Discriminative power of the filter bank is further enhanced by enforcing the features from the same category to be close to each other in the feature space, while features from different categories to be far away from each other.
    We introduce a binary selection variable vector to adaptively select what filters to share, and among what categories.
    not all the patches from the same categories are close, as they are very diverse.
    not all the local patches from different classes should be forced to be separable.
    directly learning features from image pixel values [4–9,14–18] emerges as a hot research topic in computer vision because it is able to learn data adaptive features。
    discriminative information can be critical for classification and discriminative patterns can be learned.
    By multiplying W with x, and applying an activation function F (·), we expect to generate feature f
    i = F (Wxi), which is discriminative and as compact as possible.
    We aim to minimize the distance between each feature to its positive nearest neighbours, while maximize the distance between each feature to its negative nearest neighbours
管理选项: 导出文献|

评论(0 人)

facelist doodle 涂鸦板

Copyright;  © 新科学想法 2016-2017   浙公网安备 33010202000686号   ( 浙ICP备09035230号-1 )