一种基于小波分解的快速人脸识别算法

Fast face recognition algorithm based on wavelet decomposition

提出了一种基于小波分解并结合LBP纹理特征的正交投影人脸识别算法.算法首先利用小波变换把人脸图像分解为4个不同频率和尺度的区域,然后对低频区域运用正交投影法计算相似度,对于其他3个高频区域先将其融合为一个区域,再提取LBP纹理特征向量计算相似度,最后利用动态加权自动调整低频分量和高频分量的权重得到最终识别结果.实验结果表明,该方法可以做到既利用了各分量所包含的判别信息,又将噪声、表情及光照等不利因素的影响降到最低.

An orthogonal projection for face recognition based on wavelet decomposition and LBP is proposed in this paper.Face image is firstly divided into four different frequency and scale regions using wavelet transform.Then we evaluate the similarity of the low frequency regions using orthogonal projection.The rest regions are merged into a region from which LBP vectors can be extracted for calculation.Finally,the method combines the low region similarity and the high region similarity by adjusting the weights.The experiment results demonstrate that the proposed approach not only contains the discriminant information of each component,but also reduces the influence of noise,illumination,expression,etc.