A Micro-expression Recognition System Based On CNN-CBP Features And Its Software/Hardware Co-design

Micro-expression is a kind of fleeting facial expression that people try to hide from their inner feelings,and is considered to be the most useful clues to identify lying and danger,and has an important role in understanding human deceptive behavior.Since Ekman et al.Found micro-expressions in medical video research,micro-expression studies have been conducted in a number of areas,such as clinical medicine,public safety,education,and politics.In the study of micro-expression,the study of micro-expression automatic recognition method can provide powerful tools for micro-expression research,accelerating the micro-expression principle research process and expanding micro-expression application field.In recent years,the study of micro-expression automatic recognition method is a hot spot in micro-expression research field.However,there are still many obstacles in micro-expression research,such as the problem of datasets and real-time problem.The problem of micro-expression datasets,first is that sets are not complete as the general lack of six basic expressions,followed by the amount of data relative to the expression data set is relatively small.The shortcomings of the micro-expression datasets are very unfavorable for the effective training of the model.Taking into account the micro-expression and macro-expression in the basic categories is the same,the difference lies in the duration and expression of action range,and thus the macro expression of the data set to micro-expression training is a viable option.On· the basis of[52],we use the macro expression and micro-expression datasets to train the macro expression recognition model,and then.use the macro-expression recognition model as the expression recognition module in the micro-expression recognition framework,and then use the duration of the expression to distinguish between macro-expression and micro-expression.At present,the research of micro-expression automatic recognition method is mostly based on the software level.However,in the case of human-computer interaction,interactive education and other micro-expression applications,the real-time problem of micro-expression on software is worth exploring.In order to solve the real-time problem caused by a large number of computing in micro-expression recognition,we transplanted the micro-expression automatic identification system to FPGA-based hardware and software platform,and optimized the design of system on the hardware and software platform to reduce the power consumption of the system to improve system performance with the aim of meeting the real-time premise.