基于彩色空间多特征融合的表情识别算法研究

目前的人脸表情识别方法大多是在灰度图像上采用单一特征算子,如 Local Phase Quantization(LPQ),Local Binary Patterns(LBP),Histograms Of Oriented Gradients(HOG),Gabor等,进行分类识别,但这类方法在复杂光照条件下识别率并不理想。为取得较好的识别率,本文首次提出了基于彩色图像多特征融合的表情识别算法。该算法首先在不同彩色分量上分别提取LPQ、LBP、HOG及Gabor多种特征,然后对高维特征进行线形鉴别分析并采用最近邻法进行表情分类,最后对多特征分类结果采用Adaboost算法进行融合。本文算法在具有复杂光照条件的Multi-PIE人脸库上进行了验证,取得了88.30%的平均识别率。实验结果表明：相比于基于灰度图像的单一特征识别算法,本文提出的算法能较大幅度地提高人脸表情识别率。

Facial Expression Recognition Using Multi-Feature Fusion in Color Space

Most of existing facial expression recognition algorithms apply single feature operators, such as Local Phase Quantization(LPQ),Local Binary Patterns(LBP),Histograms Of Oriented Gradients(HOG),and Gabor, but the recognition rate usually is not satisfactory under the condition of complex illumination. To improve the recognition rate, this paper proposes a new multi-feature-fusion facial expression recognition algorithm conducted in color space. Firstly, the proposed algorithm extracts multiple features, LPQ, LBP, HOG and Gabor in different color spaces, and thereafter extracts linear discriminating features and conducts classification using Nearest Neighbor Classifiers. Finally, the algorithm applies Adaboost algorithm to fuse the multiple classifiers for improving accuracy. The proposed algorithm is verified using Multi-PIE and achieves the average accuracy of 88.30%, showing that multiple feature fusion algorithm using color information largely enhances the accuracy of facial expression recognition and outperforms gray-based single feature methods evidently.