RIFeatures:旋转不变特征提取开源库 基于OpenCV

旋转不变特征

Rotation Invariant Features

一个用于计算旋转不变图像特征的小型C ++库。

一个旋转不变的基函数集

设计目标

该库的目的是从2D图像计算旋转不变特征。这些是一组特征集合,以一种对图像块方向不变的方式描述图像中的圆形图像块。因此,它们可以用于直接检测任何方向的物体。

重点是提供高效的实现:

该库能够与随机森林机器学习算法很好地集成。
仅在需要时“即时”计算所需特征,减少冗余计算。
使用傅里叶域计算来提高计算速度。
线程安全,以允许使用OpenMP多线程,避免提取特征时候数据访问出错(请注意,只有在文档中明确描述为线程安全的函数才能在并行块中使用)。
关于特征提取方法如何工作的详细描述,评估和相关科学文献的引用可以在我的博士论文中找到。你可以在我的网站上找到更多关于我如何使用该库的信息。

Objectives

The purpose of this library is to calculate rotation invariant features from 2D images. These are a set of features that describe circular image patches in an image in a way that is invariant to the orientation of the patch. They can therefore be used to detect objects in any orientation in a straightforward way.

The emphasis is on providing a highly efficient implementation that:

  • Integrates well with random forests machine learning algorithms.
  • Only calculates the required features ‘on-the-fly’.
  • Uses Fourier-domain calculations to improve calculation speeds.
  • Is thread-safe to allow learning algorithms using OpenMP multi-threading to use extract features avoiding data-races (note that only the functions that are explicitly described as thread-safe in the documentation should be used within parallel segments).

A detailed description of how the feature extraction method works, an evalution, and citations of relevant scientific literature can be found in my DPhil (PhD) thesis. You can find out more about how I am using the library on my website.

If you use this library in your research, please consider citing these papers. Other relevant publications include:

C.P. Bridge and J.A. Noble, “Object Localisation In Fetal Ultrasound Images Using Invariant Features”. Proceedings of the IEEE International Symposium on Biomedical Imaging, New York City, 2015

https://github.com/CPBridge/RIFeatures

转载请注明:《RIFeatures:旋转不变特征提取开源库 基于OpenCV

发表评论