Skip to content

AvS78/FacialMicroExpression

Repository files navigation

FacialMicroExpression

Spontaneous Facial Micro- Expression Classification. Micro-Expressions, Lie Detection, Review of Surveys, Feature extraction, HOF, LBPTOP, Facial Features, Automated feature extraction, CASMEII

Context:

Automated and reliable identification of Facial micro-expressions. Correct micro- expressions identification has several industrial uses. Today’s manual approaches have low accuracy

Aims:

Survey previous works for automated /machine learning driven facial micro-expression detection. Identify a novel alternate approach to improve Facial micro-expression detection

Methods:

Histogram of oriented optical flow (HOOF) and Local binary Pattern - Three orthogonal Planes (LBP-TOP) for feature extraction, Feed Forward Neural Network FFNN based Multi-Class Classifier. , Chinese Academy of Sciences Micro Simultaneous Expression CAMSEII for spontaneous micro-expression database to train, Metrics for Accuracy, Recall in an unbalanced dataset

Baseline Model:

Spontaneous Facial Micro-Expression Recognition using 3D Spatiotemporal Convolutional Neural Networks

Results:

In a Leave one Video Out cross validation (LOVOCV), we achieved an weighted average f1-score of 81%, precision (86%), recall (80%). The best accuracy was achieved with aa features and bb depth cc parameter FF NN. Feature extraction and modelling was done on Nvidia RTX2070 and Nvidia GTX1080 based GPU computers.

Conclusions:

We achieved satisfactory results vs the baseline. Dataset augmentation is most important activity for building Automated Spontaneous Micro-Expression detection based industrial application. Dataset augmentations requires govt authorities and research institution collaboration for an ethnically and culturally rich high quality (fps, resolution) dataset.

About

Spontaneous Facial Micro- Expression Classification

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published