Robust Facial Expression Recognition for MuCI: A Comprehensive Neuromuscular Signal Analysis

Abstract

This paper presents a comprehensive study on the analysis of neuromuscular signal activities to recognize eleven facial expressions for Muscle Computer Interfacing applications. A robust denoising protocol comprised of Wavelet transform and Kalman filtering is proposed to enhance the electromyogram (EMG) signal-to-noise ratio and improve classification performance. The effectiveness of eight different time-domain facial EMG features on system performance is examined and compared in order to identify the most discriminative one. Fourteen pattern recognition-based algorithms are employed to classify the extracted features. These classifiers are evaluated in terms of classification accuracy and processing time. Finally, the best methods that obtain almost identical system performance are compared through the Normalized Mutual Information (NMI) criterion and a repeated measure analysis of variance (ANOVA) for a statistical significant test.To clarify the impact of signal denoising, all considered EMG features and classifiers are assessed with and without this stage. Results show that: (1) the proposed denosing step significantly improves the system performance; (2) Root Mean Square is the most discriminative facial EMG feature; (3) discriminant analysis when the parameters are estimated by the Maximum Likelihood algorithm achieves the highest classification accuracy and NMI; however, ANOVA reveals no significant difference among the best methods with almost similar performance.

Realistic Facial Expression Reconstruction for VR HMD Users

“…First, we used the whole dataset and randomly selected 80% data for training and the left 20% data for testing [62], [64]. The LS-SVM with the RBF kernel was adopted for the target multiclass classifier.…”

“…Long segments can supress bias and variance of the feature, however, they may fail to reach the efficiency requirement [61]. Some recent works report that using segments with 256 msecs length is a good trade-off between the feature effectiveness and the overall processing efficiency [62], [63]. We follow the setting of [62] by segmenting the pre-processed EMG signals into non-overlapping 256-msec pieces.…”

“…With the hypothesis of the Gaussian random process, RMS provides the maximum likelihood estimation of EMG amplitude when a facial muscle is under constant force and non-fatiguing contraction. According to a recent survey [62], RMS shows superiority against the other time-domain features. We hence extract RMS from each 256-msec segment of EMG signals:…”

“…The influence of the classifier to the final prediction performance has been much studied. A recent study [62] compares 14 classifiers and reports that Least-square Support Vector Machine (LS-SVM), Regularized Discriminative Analysis (RDA), Normal Density Discriminant Function (NDDF) and Maximum-likelihood (ML) estimation provide a much higher classification accuracy against the other classifiers. ANOVA statistical analysis shows that there is no significant difference among the classification performance of the top four classifiers [64].…”

Objectively Measuring Pain Using Facial Expression: Is the Technology Finally Ready?

“…In addition, anatomical variations both in the size of faces and in the precise location of the muscles are further issues that would need to be tackled. Novel machine learning algorithms and multi-sensor arrays have improved the spatial resolution of fEMG, which partially address these technical issues [52].…”