摘要: 针对目前景象匹配算法耗时长、匹配率低等缺陷,提出基于改进的局部二进制模式(LBP)及伪Zernike矩的景象匹配算法。采用伪Zernike矩提取特征点邻域的方向和尺度信息,利用改进后的LBP算子提取邻域的纹理信息。对尺度、方向信息进行主成分分析并二值化,与纹理信息组成混合矩构成特征点邻域的特征描述子。实验结果表明,该算法的计算复杂度比其他典型算法低,匹配时间为0.05 s,基本满足实时性。给出各种情况下的匹配效果图,并将匹配率逐一对比,该算法的匹配率在标准情况下为100%,旋转变化下为64.52%,亮度变化下为53.84%,均高于其他算法,而在尺度变化下的匹配率与其他典型算法基本持平。
关键词:
景象匹配,
低维局部二进制模式,
伪Zernike,
纹理,
尺度,
方向,
亮度
Abstract: Most current scene matching algorithms have problems such as time-consuming and low matching rate. In order to solve the problems, a new scene matching algorithm based on improved Local Binary Pattern(LBP) descriptor and pseudo-Zernike is proposed. It uses pseudo-Zernike to extract information about direction and scale, and texture information is abstracted by the improved LBP descriptor. Principal component analysis is applied to scale and direction information, and binarization is put into use on principal information. Experiments turn out that the computational complexity about the algorithm is lower than the other two typical algorithms, and CPU elapsed time is 0.05 s, according with instantaneity. Matching pictures in various circumstance are also provided and compared. The algorithm’s matching rate is 100% under standard conditions, 64.52% under rotational variations, and 53.84% under brightness variations. It has the highest matching rate, and the matching rate of the algorithm is basically flat with the other algorithms.
Key words:
scene matching,
Low Demension-Local Binary Pattern(LD-LBP),
pseudo-Zernike,
texture,
scale,
orientation,
brightness
中图分类号: