Abstract: Because of the intelligence and user receptivity of mobile applications, emotion aware mobile applications are increasing. Because of the limited processing power of mobile devices, algorithms for emotion recognition on mobile devices should be implemented in real time and efficiently. This paper proposes a method for emotion recognition with high accuracy and low computational complexity in mobile applications. In this method, first, the face video is captured by a smartphone camera. Some representative frames are extracted from the video, and a human face detection module is used to extract face regions from these frames. Then, the face region is processed by Bandlet transform, and the resultant subband is divided into nonoverlapping blocks. Local binary patterns’ histograms are calculated for each block, and the histogram of all blocks is connected as a feature set to describe the facial image. Finally, Kruskal-Wallis test is used to select the most dominant features from the facial image feature sets, and these features are fed into the Gauss mixture model classifier for emotion recognition.Experimental results show that the proposed method achieves high recognition accuracy in a reasonable time.

Key words: emotion recognition, mobile applications, Bandlet transform, local binary patterns, Kruskal-Wallis test

摘要： 由于情感感知移动应用的智能性和用户易接受性，使情感感知移动应用不断增加。由于移动设备的处理能力有限，因此移动设备上的情感识别方法的算法实现应该实时和高效。提出了一个移动应用上的高精度和低计算复杂度的情感识别方法。在该方法中，人脸视频由智能手机的摄像头捕获，从视频中提取一些有代表性的帧，并且用一个人脸检测模块从这些帧中提取人脸区域。脸部区域被Bandlet变换处理，结果子波被划分为互不重叠的子块。计算每个块的局部二进制值模式的直方图，将所有块的直方图关联起来作为描述面部图像的特征集。用Kruskal-Wallis检验从面部图像特征集中选择最具优势的特征，将这些特征送入高斯混合模型分类器中进行情感识别。实验结果表明，该方法在一个合理的时间内实现了高识别精度。

关键词: 情感识别, 移动应用, Bandlet变换, 局部二进制值模式, Kruskal-Wallis检验