How to use local binary pattern to extract features?

1 次查看（过去 30 天）

显示更早的评论

Usman 2015-12-17

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/260734-how-to-use-local-binary-pattern-to-extract-features

评论： Stefan Karlsson 2015-12-17

I am new to MATLAB. Kindly help me how to use local binary pattern to extract features from images?

回答（1 个）

Image Analyst 2015-12-17

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/260734-how-to-use-local-binary-pattern-to-extract-features#answer_203573

localBinaryPattern.m

See my attached demo to compute the LBP. It doesn't extract any features from it - it merely computes it. Somehow you'd then have to analyze the LBP to get some feature that you're interested in.

7 个评论
显示 5更早的评论隐藏 5更早的评论

Image Analyst 2015-12-17

I don't know what the current best approach is. And it might depend on your field of view and how much contrast the lips have to the face. I refer you to the latest literature on it because I am not a researcher in that field of image processing: http://www.visionbib.com/bibliography/contentspeople.html#Face%20Recognition,%20Detection,%20Tracking,%20Gesture%20Recognition,%20Fingerprints,%20Biometrics

Stefan Karlsson 2015-12-17

Last I checked, LBPs where the best performing low-level features for facial, and facial parts tracking... as far as the Viola-jones cascaded ada-boost framework goes. I am not very updated though. The LBP version found in OpenCV out-performs the previous version in both efficiency and accuracy. The old version, however, was just Haar-like features, so this is not suprsing. For state of the art today, I guess deep-learning should, or will soon dominate here as well. That approach leaves feature selection in the hands of the framework.

When i used LBPs in OpenCV, the documentation mentioned their rotation invariant properties, but the actual implementation was not rotation invariant. Perhaps this has changed too now. Rotation invariance for anything with facial tracking is usually important. Scale invariance is built-into the VJ classifier, so you dont need that in the features.... good luck!

请先登录，再进行评论。

请先登录，再回答此问题。