摘要： 为提高二值描述子的分辨力和鲁棒性，提出一种新的特征组合二值描述子算法。将采样点圆形邻域划分为多个环域，通过比较任一对采样点对应环域的灰度均值获得灰度二值向量，计算采样点圆形邻域内像素点的高斯一阶梯度均值和高斯二阶偏导数均值，获取采样点对的梯度二值向量，将所有采样点对的灰度二值向量和梯度二值向量串联得到特征点的初始描述子，并采用特征筛选策略来降低描述子的维数得到低存储、强区分力的描述子。在Oxford数据集和复杂光照图片上的实验结果表明，该算法在光照变化、模糊变化和JPEG压缩条件下具有较好的鲁棒性。

关键词: 二值描述子, 特征组合, 环域, 高斯一阶梯度, 高斯二阶偏导数, 特征位

Abstract: To improve the discrimination ability and robustness of binary descriptors,we propose a new feature-combined binary descriptor algorithm.First,we divide the circular neighborhood of the sampling point into multiple ring domains,and compare the grayscale average value of corresponding ring domains of any pair of sampling points to obtain the grayscale binary vector.Then,we calculate the average values of Gaussian first-order gradient and Gaussian second-order partial derivative of all pixels in the circular neighborhood of sampling point to obtain the gradient binary vector.Next,we connect the grayscale vector and the gradient binary vector of all sampling point pairs to get the initial descriptor of the feature points.Finally,we adopt a feature selection strategy to obtain the feature bits,which are used to reduce the dimension of the descriptor to get a low-storage strong-discrimination descriptor.Experimental results on the Oxford dataset and other complex illumination images show that the proposed algorithm has good robustness under conditions of illumination variation,fuzzy variation and JPEG compression.

Key words: binary descriptor, feature-combined, ring domains, Gaussian first-order gradient, Gaussian second-order partial derivative, feature bits

中图分类号: 

• TP391.41