Deep Learning for Illumination Invariant Facial Expression Recognition

Ruiz-Garcia, Ariel; Palade, Vasile; Elshaw, Mark; Almakky, Ibrahim

doi:10.1109/ijcnn.2018.8489123

Cited by 11 publications

(5 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Autoencoders learn an encoder function f that maps an input image x to a hidden representation h = f (x), and learn a function g that maps h to a reconstruction y = g(f (x)) where y is an approximation of x. However, in recent works, [5] it has been established that the target reconstruction does not need to be the same as the input x to the autoencoder. This is supported by the theory that to be useful, an autoencoder should only learn an approximation of the target reconstruction and not an identity function that replicates it [6].…”

Section: Related Workmentioning

confidence: 99%

“…Furthermore, since GANs are known to be difficult to train due to their sensitivity to hyper-parameters and parameter initialization which often leads to mode collapse, the GASCA model is trained in a GLW fashion. However, since the greedy nature of GLW leads to error accumulation as individual layers are trained and stacked [5], we build on the gradual greedy layer-wise training algorithm from [5] and adapt it for adversarial autoencoders. Accordingly, we introduce the GAN gradual greedy layer-wise (GANGGLW) training framework and formally define it in Algorithm 1.…”

Section: Generative Adversarial Stacked Convolutional Autoencodersmentioning

confidence: 99%

“…For the GASCA 2 model we incorporate illumination invariance normalization training by taking images with disproportionate degrees of illumination and reconstructing them as images with good illumination: good illumination is determined by their relative luminance Y as done by [5].…”

Section: Unsupervised Feature Learningmentioning

confidence: 99%

See 2 more Smart Citations

Generative Adversarial Stacked Autoencoders for Facial Pose Normalization and Emotion Recognition

Ruiz-Garcia¹,

Palade²,

Elshaw³

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

In this work, we propose a novel Generative Adversarial Stacked Autoencoder that learns to map facial expressions with up to ±60 degrees to an illumination invariant facial representation of 0 degrees. We accomplish this by using a novel convolutional layer that exploits both local and global spatial information, and a convolutional layer with a reduced number of parameters that exploits facial symmetry. Furthermore, we introduce a generative adversarial gradual greedy layer-wise learning algorithm designed to train Adversarial Autoencoders in an efficient and incremental manner. We demonstrate the efficiency of our method and report state-of-the-art performance on several facial emotion recognition corpora, including one collected in the wild.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Generative Adversarial Stacked Convolutional Autoencodersmentioning

confidence: 99%

See 1 more Smart Citation

Generative Adversarial Stacked Autoencoders for Facial Pose Normalization and Emotion Recognition

Ruiz-Garcia¹,

Palade²,

Elshaw³

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…Ruiz-Garcia et.al [11] proposed a deep Stacked Convolutional Autoencoder (SCAE) method. It is trained to reconstruct the images with different illumination to the mean luminance.…”

Section: Related Workmentioning

confidence: 99%

“…Therefore, the main aim of facial expression recognition methods and approaches is to enable machines to automatically estimate the emotional content of human face. In intelligent tutoring systems, emotions and learning are inextricably bound together; so recognizing learners' emotional states could significantly improve the efficiency of the learning procedures delivered to them [9]- [11].…”

Section: Introductionmentioning

confidence: 99%

Illumination Invariant Facial Expression Recognition using Convolutional Neural Networks

Rao*¹,

Rao²

2019

IJRTE

View full text Add to dashboard Cite

In this work, we propose a prospective novel method to address illumination invariant system for facial expression recognition. Facial expressions are used to convey nonverbal visual information among humans. This also plays a vital role in human-machine interface modules that have invoked attention of many researchers. Earlier machine learning algorithms require complex feature extraction algorithms and are relying on the size and uniqueness of features related to the subjects. In this paper, a deep convolutional neural network is proposed for facial expression recognition and it is trained on two publicly available datasets such as JAFFE and Yale databases under different illumination conditions. Furthermore, transfer learning is used with pre-trained networks such as AlexNet and ResNet-101 trained on ImageNet database. Experimental results show that the designed network could recognize up to 30% variation in the illumination and it achieves an accuracy of 92%.

show abstract