为了提取复杂环境下人脸图像的有效特征,提出了一种结合DMMA(discriminative multi-manifold analysis)和方向梯度直方图(HOG)特征提取算法,利用了一种新的自适应方法计算子图像块的相似度。在DMMA算法中,将一幅样本图像分为不重叠的子图像块后,对每一个小块使用HOG算子进行处理,处理后形成一个统计流形,然后进行特征提取,利用基于重建的流形—流形间的距离最近邻方法进行分类识别。在AR人脸库和FERET人脸库上的实验结果表明,该算法对人脸图像的光照和几何变化比传统的DMMA算法识别性能更好。%In order to extract effective features of the complex environment face image,this paper presented a novel method by fusing HOG features and discriminative multi-manifold analysis (DMMA)features.It applied a new adaptive method to calcu-late similarity between patches of the face image.First,it partitioned each face image into several nonoverlapping patches to form an image set for each sample per person.Then it used histogram of the oriented gradient (HOG)operator to extract image histogram of each an image set.The histogram of each an image set formed a statistics manifold.Last it applied DMMA algo-rithm to obtain the low-dimensional face image feature.It used the reconstruction-based manifold-manifold distance to identify the unlabeled subjects.Experimental results show that the algorithm for face images of light and geometry changes is superior to the general recognition DMMA algorithms on the AR database and FERET database.
展开▼