Micro-expression Recognition Method View Deep Learning

Micro-expression as an effective nonverbal clue,has the typical characteristics of short duration and small amplitude.which has broad application prospects in various fields.In view of the difficulty of micro-expression recognition,the method of micro-expression detection and the algorithm of micro-expression recognition model by deep learning were proposed.The main work as follows:First of all,according to the micro-expression model’s requirements for input data,the appropriate micro-expression datasets were selected,and the preprocessing of micro-expression datasets were completed through the process of micro-expression data screening,unified division of micro-expression states,the data fusion of different datasets,data types equalization,apex frame positioning,image frame cropping,and feature encoding.The preprocessing provides key elements and core indicators for micro-expression recognition.Secondly,for the problem of micro-expression location in video,a micro-expression detection method based on pixel difference was proposed.A long video dataset that contains two types of expressions at the same time was chosen,and the facial areas involved in the generation of the two expressions were summarized,then introduce the theoretical basis and calculation process of the pixel difference in detail.Through analysis the change of pixel difference in a long video sequence,the motion threshold and time of micro expression action were summarized to complete the micro expression localization and detection.Finally,in view of the unsatisfactory effect of micro-expression recognition,a micro-expression recognition algorithm based on CBAM-DPN network based on deep learning was proposed.The algorithm integrates the CBAM module after the Bottleneck layer of the original DPN network.On the basis of the dual-path,the multi-dimensional attention mechanism was added,which can better pay attention to the area that micro-expression actions need to pay attention.Through experimental comparison and analysis,the CBAM-DPN algorithm improves the network’s ability to extract features at the cost of a very small network overhead,and provides a better recognition effect.After the micro-expression is divided into risk levels by the expert scoring method,the identification model can be used to obtain the probability that the micro-expression belongs to each risk level.The probability was used as the input of the risk value calculation formula to complete the personnel safety analysis and provide suggestions for airports and airlines.