Keywords
Deep Learning, Neural Network, Computer Vision
Select the category the research project fits.
Mathematical/Computational Sciences
Is this submission part of ICaP/PW (Introductory Composition at Purdue/Professional Writing)?
No
Abstract
Emotion recognition recently becomes a popular topic of machine learning and computer vision and generates a wide range of applications in other academic fields as well as in our everyday life. The primary idea is to process the input of a human facial expression frame and output a result that classifies such expression into some designated emotions such as basic seven emotions or other compound emotions.The existing conventional method requires a handcrafted feature extractor of facial Action Units(AUs) to extract feature from designated Facial Landmark regions, and these extracted AUs codes are processed through traditional machine learning algorithm such as Nearest Neighbors and SVM, which is a typical type of linear classifier. The problem with conventional method is that the lighting variations and different position of object may corrupt the feature vector so that the accuracy is greatly reduced. In this project, we approach the problem by applying Convolutional Neural Networks(CNNs), which does not require the step of handcrafted feature extraction in some fixed designated facial landmark region but produces an system that automatically extracts feature segments and completes classification process by feedforward calculations . This system achieves the relatively most optimal solution through the process of backpropagation in which the algorithm learns the weights through stochastic gradient descent that can find the directions that best minimize the loss from the ground truth. The numerical result of the algorithm will show a probabilistic result of each labeled class. Furthermore, such method best resolves the issues of lighting variations and different orientation of object in the image and thus achieves a higher accuracy.
Recommended Citation
Li, Chieh-En James and Zhao, Lanqing, "Emotion Recognition using Convolutional Neural Networks" (2019). Purdue Undergraduate Research Conference. 63.
https://docs.lib.purdue.edu/purc/2019/Posters/63
Emotion Recognition using Convolutional Neural Networks
Emotion recognition recently becomes a popular topic of machine learning and computer vision and generates a wide range of applications in other academic fields as well as in our everyday life. The primary idea is to process the input of a human facial expression frame and output a result that classifies such expression into some designated emotions such as basic seven emotions or other compound emotions.The existing conventional method requires a handcrafted feature extractor of facial Action Units(AUs) to extract feature from designated Facial Landmark regions, and these extracted AUs codes are processed through traditional machine learning algorithm such as Nearest Neighbors and SVM, which is a typical type of linear classifier. The problem with conventional method is that the lighting variations and different position of object may corrupt the feature vector so that the accuracy is greatly reduced. In this project, we approach the problem by applying Convolutional Neural Networks(CNNs), which does not require the step of handcrafted feature extraction in some fixed designated facial landmark region but produces an system that automatically extracts feature segments and completes classification process by feedforward calculations . This system achieves the relatively most optimal solution through the process of backpropagation in which the algorithm learns the weights through stochastic gradient descent that can find the directions that best minimize the loss from the ground truth. The numerical result of the algorithm will show a probabilistic result of each labeled class. Furthermore, such method best resolves the issues of lighting variations and different orientation of object in the image and thus achieves a higher accuracy.