基于积分投影和LSTM的微表情识别研究

Research on micro-expression recognition based on integral projection and LSTM

现有的微表情识别研究主要是利用基于局部二值模式(LBP)改进的算法并结合支持向量机(SVM)来识别。最近,积分投影开始应用于人脸识别领域。长短时记忆网络(LSTM)作为循环神经网络,可以用来处理时序数据。因此提出了结合积分投影和LSTM的模型(LSTM-IP),在最新的微表情数据库CASME II上进行实验。通过积分投影得到水平和垂直投影向量作为LSTM输入并分类,同时采用了防止过拟合技术。实验结果表明,LSTM-IP算法取得了比以前的方法更好的精度。

The existing research on micro expression recognition is mainly based on the improved LBP （local binary patterns） algorithm and SVM （support vector machine）. Recently, integral projection has been applied in the field of face recognition. The long and short memory network （LSTM）, as a kind of recurrent neural network, can be used to process time series data. So LSTM-IP model, which combines integral projection with LSTM, is proposed, and experimented on the latest micro-expression database CASME II. The horizontal and vertical projection vectors obtained by integral projection are used as the input of LSTM and classified, and the over-fitting preventing method is used. The experimental results show that LSTM-IP algorithm gets better results than the previous method.