一种基于特征融合降维的表情识别算法

Expression Recognition Based on Feature Fusion Method for Dimensionality Reduction

为了降低人脸表情识别过程中特征分类的计算量,采用了一种基于特征融合降维的表情识别算法。该算法首先对表情图像进行预处理,再利用Gabor小波多尺度多方向的特性对图像进行滤波,针对同一尺度下8个不同方向的几幅特征图像,对其中特征值最大的图像编码作为新特征图像的像素值,此时特征图像的维数降为原来的1/8。最后利用统计直方图对融合后的特征图像进行分块特征统计,将统计信息作为最终的特征信息进行分类。实验结果表明,该方法在保证人脸表情识别率的前提下减少了特征图像的计算量,提高了系统效率。

In order to reduce the dimensionality and the calculation of expression recognition,a new method based on the traditional framework is used.Firstly,features of expression images were extracted with Gabor wavelet of the preprocessed images.Then,getting the number of the image which has the largest pixel value as the new feature for the eight different orientations on the same scale.So the dimension of characteristic will be reduced to 1/8of the original.Finally,counting the feature of fused image which has divided into blocks by statistical histogram.And treating the statistical information as the final feature for classification.Experimental results show that the method reduces the computational of classification of the images in the context of guaranteeing facial expression recognition rate and also improves the system efficiency.