Research On Micro-expression Recognition Based On Euler's Video Zooming Algorithm

Micro-expression is one of the important ways to transmit emotions.But micro-expressions are highly disguised.Compared with the normal expressions that people deliberately show,the most realistic feelings and motivations can be reflected through micro-expressions in human beings.With the deep exploration of lies detection,there is more and more attention payed by researchers in micro-expression recognition,especially in automatic way.The results of this research can applicate in lots of aspects such as homeland security,judicial investigation,clinical medicine and so on.However,the key points in most of the research presently is that to analyze the spatial features of micro-expressions,or to process the information of micro-expressions in the time dimension.As a manifestation of emotion,there is a certain connection between the changes of heart rate and micro-expression.Based on the factors of heart rate,we analyze the relationship between heart rate changes and micro-expressions,then doing studies whether the capture of heart rate changes can optimize the automatic recognition algorithm for micro-expressions in this paper.To achieve heart rate extraction in non-contact way,we use Eulerian Video Magnification algorithm to amplify the signal in a specific region of each frames in a video,and then process the signal by filter that we designed.Finally calculate the real-time heart rate of people who in the video.We analyzed the micro-expression database which published already and summarized the advantages and applicability of each database.The CAS(ME)^2 micro-expression database was selected as the data source for this study.Heart rates of data in the database which we choose are extracted by the heart rate extraction algorithm.Matching the connection between heart rate changes and the database information,so that we can form a benchmark for the final micro-expression recognition.Due to the aspect of that data source which are the dynamic video and keeping conditions uniform,we use the LBP-TOP feature extraction method and SVM classification algorithm to recognize micro-expression.We combined the methods of heart rate extraction and the traditional micro-expression recognition algorithm.Through the cross-combination of samples which are the sample of heart rate variation and the sample of original database,and recognizing the micro-expression which contain in the sample of heart rate changes,we can analysis the optimization of micro-expression recognition that based on the heart rate change.The experimental results show that Eulerian Video Magnification algorithm can be well applied to heart rate extraction.95%accuracy can be achieved through the heart rate detection by using the test database which are videos that be took in different time and different volunteers.The real-time heart rate is very smooth.Under the database of CAS(ME)^2,there is a certain matching relationship between emotions and heart rate changes such as fear,anger,sadness and so on.In the part of micro-expression recognition,under the same conditions of database,the feature extraction algorithm and the classification algorithm,we found that by add the factor of heart rate to the micro-expression recognition,the accuracy maintains a high degree of similarity.One set of data is even 0.01%higher than the results of the original recognition method.Thus,it can be shown that there is a great relationship between two dynamic changes which are heart rate and micro-expression.