Recognizing Spontaneous Facial Expressions using Deep Convolutional Neural Networks
(2018) In Master's Theses in Mathematical Sciences FMAM05 20181Mathematics (Faculty of Engineering)
- Abstract
- Emotion recognition is a relatively new research area within image analysis where machine learning models learn to interpret facial expressions. In this thesis paper the authors have investigated how close to human accuracy selected deep convolutional neural network models can reach on the Emotion Recognition Problem, what the purpose of such an application might be and what compromises can be made when balancing computational power and classification performance.
A number of deep convolutional neural network models of varying depth were trained on mainly two datasets; FER2013 and AffectNet. Results showed that the best model managed to reach 74.6 % accuracy on a three label testset, where a human performed 80.7 %. The conclusion was... (More) - Emotion recognition is a relatively new research area within image analysis where machine learning models learn to interpret facial expressions. In this thesis paper the authors have investigated how close to human accuracy selected deep convolutional neural network models can reach on the Emotion Recognition Problem, what the purpose of such an application might be and what compromises can be made when balancing computational power and classification performance.
A number of deep convolutional neural network models of varying depth were trained on mainly two datasets; FER2013 and AffectNet. Results showed that the best model managed to reach 74.6 % accuracy on a three label testset, where a human performed 80.7 %. The conclusion was made that for the most interesting use cases, label binning facial expressions into three classes positive, neutral and negative might be favorable. Lastly, as for trade-off when balancing accuracy and computational power, deeper networks perform insignificantly better than the more shallow ones, while the latter dramatically reduce the computational effort. (Less) - Popular Abstract (Swedish)
- Datorer blir allt snabbare och mer kraftfulla, men kan de klara av att lösa problem som anses kräva "mänsklig" intelligens, så som att tolka ansiktsuttryck? Det visar sig att så faktiskt är fallet - en dator kan nå nästan samma noggrannhet som en människa.
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/8943330
- author
- Söderberg, Erik LU and Jönsson, Andreas LU
- supervisor
-
- Karl Åström LU
- Martin Ahrnbom LU
- organization
- course
- FMAM05 20181
- year
- 2018
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- facial expression recognition, deep neural network, machine learning, image analysis
- publication/series
- Master's Theses in Mathematical Sciences
- report number
- LUTFMA-3348-2018
- ISSN
- 1404-6342
- other publication id
- 2018:E25
- language
- English
- id
- 8943330
- date added to LUP
- 2018-05-31 14:30:04
- date last changed
- 2018-05-31 14:30:04
@misc{8943330, abstract = {{Emotion recognition is a relatively new research area within image analysis where machine learning models learn to interpret facial expressions. In this thesis paper the authors have investigated how close to human accuracy selected deep convolutional neural network models can reach on the Emotion Recognition Problem, what the purpose of such an application might be and what compromises can be made when balancing computational power and classification performance. A number of deep convolutional neural network models of varying depth were trained on mainly two datasets; FER2013 and AffectNet. Results showed that the best model managed to reach 74.6 % accuracy on a three label testset, where a human performed 80.7 %. The conclusion was made that for the most interesting use cases, label binning facial expressions into three classes positive, neutral and negative might be favorable. Lastly, as for trade-off when balancing accuracy and computational power, deeper networks perform insignificantly better than the more shallow ones, while the latter dramatically reduce the computational effort.}}, author = {{Söderberg, Erik and Jönsson, Andreas}}, issn = {{1404-6342}}, language = {{eng}}, note = {{Student Paper}}, series = {{Master's Theses in Mathematical Sciences}}, title = {{Recognizing Spontaneous Facial Expressions using Deep Convolutional Neural Networks}}, year = {{2018}}, }