Spontaneous Smile Detection with Application of Landmark Points Supported by Visual Indications

David C. Wyld et al. (Eds) : CSITA, ISPR, ARIN, DMAP, CCSIT, AISC, SIPP, PDCTA, SOEN - 2017
pp. 187– 194, 2017. © CS & IT-CSCP 2017 DOI : 10.5121/csit.2017.70118
SPONTANEOUS SMILE DETECTION WITH
APPLICATION OF LANDMARK POINTS
SUPPORTED BY VISUAL INDICATIONS
Karolina Nurzynska and Bogdan Smolka
Faculty of Automatic Control, Electronics, and Computer Science
Silesian University of Technology,
ul. Akademicka 2A, 44-100 Gliwice, Poland
Karolina.Nurzynska@polsl.pl, Bogdan.Smolka@polsl.pl
ABSTRACT
When automatic recognition of emotion became feasible, novel challenges has evolved. One of
them is the recognition whether a presented emotion is genuine or not. In this work, a fully
automated system for differentiation between spontaneous and posed smiles is presented. This
solution exploits information derived from landmark points, which track the movement of
fiducial elements of face. Additionally, the smile intensity computed with SNiP (Smile-Neutral
intensity Predictor) system is exploited to deliver additional descriptive data. The performed
experiments revealed that when an image sequence describes all phases of smile, the landmark
points based approach achieves almost 80% accuracy, but when only onset is exploited,
additional support from visual cues is necessary to obtain comparable outcomes.
KEYWORDS
Smile veracity recognition, Landmarks, Smile intensity, Classification
1. INTRODUCTION
Automatic analysis of image content finds its application in many various domains: starting from
medical image analysis in order to facilitate diagnosis, going through detection of flaws in
products on a production line in a factory, and finishing on human behaviour interpretation. In all
these situations, similar approaches for image processing and recognition are implemented. The
image content is explored in order to derive some characteristics, which later on become a feature
set. These characteristics may originate from visual information, when for instance the texture is
exploited as a data source, or describe abstract details, when shape and active shape features are
considered.
Having an image sequence of facial gesture it is possible to recognize which emotion was
presented with very high efficiency, as was proved in several research works [2, 3, 4, 5, 6].
However, the accuracy of such system performance decreases significantly, when real life
scenarios are considered [7, 8, 9], because programs trained on posed emotions displayed in
laboratory environment are used to very strong facial gesture presentation, which is not common

188 Computer Science & Information Technology (CS & IT)
in daily life, not to mention lighting variation or occlusions. Nevertheless, in most cases
recognition of the happiness emotion is on satisfactory level. Therefore, further investigations
were performed, which aimed at recognition whether the smile corresponding to this emotion is a
genuine one or not [10, 11, 12].
When designing a system for smile veracity recognition, two most common approaches are
identified: One exploits facial landmarks for feature calculation [10], where distances and
characteristic points movement relationship are considered. The other uses data sampled from the
image texture [12] and combines them with smile intensity measure. The presented solution is a
combination of both. The veracity of smile is described by automatically derived characteristic
points, which are later normalized before the features are computed. On the other hand, a smile
intensity function derived from the image textural content, following the idea of SNIP system [1],
is applied.
The paper starts with description of facial landmarks determination presented in Sec. 2. Next,
Sec. 3 presents the method for smile intensity function calculation. The details of image sequence
corpora selected for results verification is given in Sec. 4. Then, the results are presented in Sec. 5
and conclusions are drawn in Sec. 6.
2. CHARACTERISTIC POINTS
In order to describe an emotion presented on image, the positions of facial landmarks called also
characteristic points are calculated. These landmarks correspond to characteristic regions of the
face, such as eyes, nose, lips, etc. and are exploited to track their changes during the presentation
of emotion. Figure 1 depicts the placement of data obtained with implementation presented in
[13].
2.1 Data Normalization

Computer Science & Information Technology (CS & IT) 189
Figure 1: Characteristic points denoted on the facial image.
2.2 Feature Calculation

Figure 2. Smile intensity magnitude
Figure 3: Exemplary frames from image sequence in UvA-NEMO dataset.
3. SMILE INTENSITY

4. DATABASE
5. EXPERIMENTS AND RESULTS
Figure 4: Influence of facial region described by features on classification rates.
Figure 5: Classification rates when early fusion method was chosen for classification.

Results gathered in Fig. 4 revealed that it is possible to distinguish the posed from genuine smile
with high accuracy. This plot shows also that the biggest influence on smile type detection has the
onset phase, however for data exploiting smile intensity, similar results were obtained for the
apex. Moreover, incorporating information obtained in other phases in all feature vector proved to
give the best results. As presumed the global approach outcomes are slightly worse and its
combination with all does not improve the accuracy (total), except for the smile intensity case. It
is also worth noticing that it is easier to distinguish spontaneous emotion from posed ones, when
lips magnitude is considered.
In order to create more accurate classifier, several options were considered. Figure 5 collects
correct classification performances for concatenated features vectors describing all possible
combinations of pairs of accessible descriptors. The best score of 79% was obtained for the total
feature vector build from eyelids and lip corner amplitudes, and slightly outperformed the results
when data was described by lips corners supported by smile intensity, which reached 78.47%.
Finally, Fig. 6 presents correct classification performance ratios when two different approaches fo
outcome computation were investigated. The early fusion data was obtained by concatenation of
feature vectors computed in first experiment and then classified with SVM. The late fusion
applied three SVMs classifiers for each feature vector generated in first experiment and used
voting to obtain the final score.
The conducted experiments show that it is possible to determine with high probability the
veracity of smile using automatic method for landmark detection or visual approach of smile
intensity description. Exploiting data derived only from landmark points with calculation of
features both for local phases supported with global information enabled correct recognition with
almost 80% accuracy. On the other hand, when only onset phase was accessible, combination of
amplitude data derived from landmarks with those computed when visual data was exploited
were not far behind.
Figure 6: Classification rates when different fusion methods were chosen for classification.

6. CONCLUSIONS
This work presents fully automatic approach to smile veracity detection, which is based on
landmarks movements estimation within an image sequence and is supported by information
derived from visual cues. Namely, the lip corner and eyelid movements magnitudes are derived
from characteristic point location and smile intensity function is achieved from the SNiP system.
The feature vector was obtained as a combination of parameters computed for each input
function. Using SVM for classification of parameters derived from these data, it was shown, that
it is possible to achieve the accuracy of almost 80%, when all information was used. However,
only slight deterioration is noticed, when parameters were reduced to describe the onset phase
only. In further research, the visual content of images will be explored to improve the
classification performance.
ACKNOWLEDGMENT
This work was supported by the Polish National Science Centre (NCN) under the Grant: DEC-
2012/07/B/ST6/01227. K. Nurzynska was partially supported by statutory funds for young researchers
(BKM/507/RAU2/2016) of the Institute of Informatics, Silesian University of Technology, Poland. B.
Smolka received partial funding from statutory funds (BK/213/RAU1/2016) of the Institute of
Automatic Control, Silesian University of Technology, Poland.
REFERENCES
[1] K. Nurzynska and B. Smolka, Computational Vision and Medical Image Processing V, ch. SNIP:
Smile–Neutral facial display Intensity Predictor, pp. 347–353. Taylor & Francis Group, 2016.
[2] K. Mase, “An application of optical flow - extraction of facial expression,” in IAPRWorkshop on
Machine Vision Applications, pp. 195–198, 1990.
[3] J. Cohn, A. Zlochower, J. Lien, and T. Kanade, “Automated face analysis by feature point tracking
has high concurrent validity with manual facs coding,” Psychophysiology, vol. 36, no. 2, pp. 35–43,
1999.
[4] J. Cohn, A. Zlochower, J. Lien, and T. Kanade, “Feature-point tracking by optical flow discriminates
subtle differences in facial expression,” in Proceedings of the 3rd IEEE International Conference on
Automatic Face and Gesture Recognition (FG ’98), pp. 396 – 401, April 1998.
[5] B. Fasel and J. Luettin, “Automatic facial expression analysis: a survey,” Pattern Recognition, vol.
36, no. 1, pp. 259 – 275, 2003.
[6] C. Shan, S. Gong, and P. W. McOwan, “Facial expression recognition based on local binary patterns:
A comprehensive study,” Image Vision Comput., vol. 27, pp. 803–816, May 2009.
[7] A. Dhall, R. Goecke, J. Joshi, M. Wagner, and T. Gedeon, “Emotion recognition in the wild challenge
(EmotiW) challenge and workshop summary,” in Proceedings of the 15th ACM on International
Conference on Multimodal Interaction, ICMI ’13, (New York, NY, USA), pp. 371–372, ACM, 2013.
[8] J. M. Girard, J. F. Cohn, L. A. Jeni, M. A. Sayette, and F. De la Torre, “Spontaneous facial expression
in unscripted social interactions can be measured automatically,” Behavior Research Methods, vol.
47, no. 4, pp. 1136–1147, 2015.

[9] J. Chen, Y. Ariki, and T. Takiguchi, “Robust facial expressions recognition using 3D average face
and ameliorated AdaBoost,” in Proceedings of the 21st ACM International Conference on
Multimedia, MM ’13, (New York, NY, USA), pp. 661–664, ACM, 2013.
[10] H. Dibeklio˘glu, A. A. Salah, and T. Gevers, Are You Really Smiling at Me? Spontaneous versus
Posed Enjoyment Smiles, pp. 525–538. Berlin, Heidelberg: Springer Berlin Heidelberg, 2012.
[11] H. Dibeklio˘glu, A. A. Salah, and T. Gevers, “Recognition of genuine smiles,” IEEE Transactions on
Multimedia, vol. 17, pp. 279–294, March 2015.
[12] M. Kawulok, J. Nalepa, K. Nurzynska, and B. Smolka, “In search of truth: Analysis of smile intensity
dynamics to detect deception,” in Advances in Artificial Intelligence - IBERAMIA 2016 - 15th Ibero-
American Conference on AI, San José, Costa Rica, November 23-25, 2016, Proceedings, pp. 325–
337, 2016.
[13] A. Asthana, S. Zafeiriou, S. Cheng, and M. Pantic, “Incremental face alignment in the wild,” in 2014
IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2014, Columbus, OH, USA,
June 23-28, 2014, pp. 1859–1866, 2014.
[14] C.-C. Chang and C.-J. Lin, “LIBSVM: A library for support vector machines,” ACM Transactions on
Intelligent Systems and Technology, vol. 2, pp. 27:1–27:27, 2011. Software available at
http://www.csie.ntu.edu.tw/~cjlin/libsvm.
AUTHORS
Karolina Nurzynska received her M.E. and Ph.D. degree in the field of
computer science from the Silesian University of Technology, Poland in 2005 and
2009, respectively. From 2009 to 2011 she was a Postdoc Researcher in the
School of Electrical and Computer Engineering at the Kanazawa University,
Japan. From 2011 to 2014 she was leading a project concerning visualisation of
underground coal gasification in Central Mining Institute in Katowice, Poland,
where in 2012 she was appointed the Assistant Professor position. Since 2013 she
is an Assistant Profesor at the Department of Automatic Control, Electronics, and
Computer Science of the Silesian University of Technology, Poland. Her research
interests include image processing and understanding, data classification and 3D surface reconstruction.
Bogdan Smolka received the Diploma in Physics degree from the Silesian
University, Katowice, Poland, in 1986 and the Ph.D. degree in computer science
from the Department of Automatic Control, Silesian University of Technology,
Gliwice, Poland, in 1998. From 1986 to 1989, he was a Teaching Assistant at the
Department of Biophysics, Silesian Medical University, Katowice, Poland. From
1992 to 1994, he was a Teaching Assistant at the Technical University of
Esslingen, Germany. Since 1994, he has been with the Silesian University of
Technology. In 1998, he was appointed as an Associate Professor in the
Department of Automatic Control. He has also been an Associate Researcher with
the Multimedia Laboratory, University of Toronto, Canada since 1999. In 2007, Dr. Smolka was promoted
to Professor at the Silesian University of Technology. He has published over 300 papers on digital signal
and image processing in refereed journals and conference proceedings. His current research interests
include low-level color image processing, human-computer interaction and visual aspects of image quality.

Spontaneous Smile Detection with Application of Landmark Points Supported by Visual Indications

More Related Content

What's hot (20)

Similar to Spontaneous Smile Detection with Application of Landmark Points Supported by Visual Indications (20)

More from cscpconf (20)

Recently uploaded (20)

Spontaneous Smile Detection with Application of Landmark Points Supported by Visual Indications