Facial Expression Recognition Based on Region-Wise Attention and Geometry Difference
Facial expression is usually considered as a face movement process.People can easily distinguish facial expressions via subtle facial changes.Inspired by this,we design two models that are expected to better recognize facial expressions by capturing subtle changes in the face.First,we consider to re-calibrate the response of different facial regions to highlight several special facial areas.According to this idea,we constructed cross-channel region-wise attention network(CCRAN),which can underline the important information and mine the correlations between different facial regions effectively.Moreover,we use the feature subtraction method to obtain geographical facial difference information.Based on this idea,we constructed temporal geometric frame difference network(TGFDN),which accepts the facial landmark points as input.These points are extracted from the facial expression frames.This network can effectively extract the slight changes of geographical information on the expression sequences.Through properly fusing these two networks,we have achieved competitive results on the CK+and Oulu-CASIA databases.
Facial expression recognition Attention mechanisms Temporal difference
Heran Du Huicheng Zheng Mingjing Yu
School of Data and Computer Science,Sun Yat-sen University,Guangzhou,China;Key Laboratory of Machine Intelligence and Advanced Computing,Ministry of Education,Guangzhou,China;Guangdong Key Laboratory of Information Security Technology,135 West Xingang Road,Guangzhou 510275,China
国际会议
广州
英文
183-194
2018-11-23（万方平台首次上网日期，不代表论文的发表时间）