Locality-preserving descriptor for robust texture feature representation
详细信息 查看全文
- 作者：Jongbin RyuAuthor Vitae ; Hyun S. YangAuthor Vitae
- 关键词：Texture classification ; Rotation invariant feature ; Texture feature description ; Locality preserving descriptor (LPD)
- 刊名：Neurocomputing
- 出版年：2016
- 出版时间：19 November 2016
- 年：2016
- 卷：214
- 期：Complete
- 页码：729-738
- 全文大小：1844 K
文摘
Recent texture classification methods include rotation-invariant feature-encoding procedures based on local binary patterns. Such methods are robust to rotational changes, but they result in discarded locality information (i.e., geometrical information) of texture images. Hence, we present a locality-preserving descriptor that encodes rotation-invariant features. The proposed method samples neighboring pixels similar to procedures based on local binary patterns, but neighborhood sampling is done based on Gabor Maximum Orientation to ensure that locality is preserved. Conventional methods discard the locality information because bit patterns are grouped for rotation-invariant encoding. In other words, the grouping procedure ignores the geometry of each bit pattern. However, the proposed locality-preserving descriptor does not include a grouping procedure, though it is both rotation invariant and locality preserving, owing to Gabor filter banks. In the experiments, we demonstrated the state-of-the-art performance of our method with widely used texture datasets.