Learning Pain from Emotion: Transferred HoT Data Representation for Pain Intensity Estimation

Abstract

Automatic 1 monitoring for the assessment of pain can significantly improve the psychological comfort of patients. Recently introduced databases with expert annotation opened the way for pain intensity estimation from facial analysis. In this contribution, pivotal face elements are identified using the Histograms of Topographical features (HoT) which are a generalization of the topographical primal sketch. In order to improve the discrimination between different pain intensity values and respectively the generalization with respect to the monitored persons, we transfer data representation from the emotion oriented Cohn-Kanade database to the UNBC McMaster Shoulder Pain database.

Cumulative attributes for pain intensity estimation

“…This is in line with observations in [8,12,17] where such techniques have been shown to yield performance gains. Among the individual features, the HOG descriptor gives the best performance with the geometric features performing much worse.…”

“…The face is first divided into a uniform grid of cells, from which Local binary patterns are extracted over a specified time window. Florea et al [12] proposed a histogram of topographical features for pain estimation in a transfer learning framework. Exploiting the temporal progression of pain expression from neutral through the apex and then back to neutral, Zhong et al [45] propose ordinal information for pain estimation.…”

“…First, we obtain 66 facial points for each face image using the publicly available iCCR landmark tracker [34]. This is in contrast to previous work [8,12,17,18,33] which use the manually annotated facial points provided with the database. Facial points obtained from the tracker are used for face registration.…”

Recurrent Convolutional Neural Network Regression for Continuous Pain Intensity Estimation in Video

“…Pain levels Measures Classifier Cross Validation C-APP + S-PTS [2] C-2 OPI, PSPI SVM Leave One Subject Out PTS + APP [22] C-2 PSPI SVM Leave One Subject Out PTS, APP [21] C-2 PSPI SVM Leave One Subject Out SAPP +SPTS + CAPP [22] C-2 PSPI SVM+LLR Leave One Subject Out AAM [24] C-2 OPI, PSPI SVM Leave One Subject Out PLBP, PHOG [19] C-2 PSPI SVM 10-fold Auto Encoder [27] C-2 PSPI SVM Leave One Subject Out TPS [32] C-2 PSPI DML + SVM Leave One Subject Out Canny Edge [15] C-2/C-8 OPI, PSPI TBM 3-fold LBP [6] C-2 PSPI Transfer Learning Leave One Subject Out PCA [1] C-3 VAS SVM, Angular Distance 10-fold DCT + LBP [18] R PSPI RVR Leave One Subject Out Hess + Grad + AAM [11] R PSPI SVM Leave One Subject Out 2Standmap [16] R PSPI RVR Leave One Subject Out the most successful applications refer to sequential tasks such as [13], [14]. RNN has been characterized by connecting hidden layers of the current time step and several previous time steps.…”

Late fusion of facial dynamics for automatic expression recognition

“…Automatic facial expression recognition is the fundamental for multiple applications in various domains such as security, computer science, education, automotives [2], crime investigation [3], interactive gaming [4], health support appliances, research of depression [5], or pain detection [6]. However, the current state of the art indicates that only particular solutions have reached maturity and further advance is still required.…”