Contourlet变换域下的目标特征提取与识别

Feature Extraction and Recognition of Object Based on Contourlet transform

利用Contourlet变换对于高维信号的表示能力,在Contourlet变换下提取不变矩特征以及局部Contourlet二值模式特征,通过特征组合,提出了一种在多种外界变化条件下都具有较好稳定性的目标特征提取技术。对于Contourlet分解的低频分量,计算多尺度自卷积矩不变特征;对于Contourlet分解的高频分量,计算其局部Contourlet二值模式(LCBP),并利用两状态HMT描述LCBP系数,得到LCBP-HMT模型,提取模型参数作为特征向量;最后将提取出的低频特征以及高频统计特征组合成特征向量,从而结合了MSA的全局不变性以及LCBP的多尺度、多方向局部描述特性。最后分别对目标的二值图像和灰度图像进行实验,证明了算法在各种变化条件下均具有较好的识别效果。

A new method of feature extraction and recognition of object based on multiscale geometric analysis is proposed. We combined the low frequency of contourlet transform with multiscale autoconvalution. For the high frequency of contourlet transform,we compute the local contourlet binary patterns and the LCBP coefficients are modeled by a two-state HMT. The parameters of LCBP-HMT model are extracted as features vector. At last,the low frequency features and high frequency statistic features together as features vector are combined. Thereby the global invariance of MSA and the multiscale and multidirectional partial characterization of LCBP are combined. Experimental results of binary images and intensity images of aircraft show that the algorithm has a good performance of recognition.