Claim Missing Document
Check
Articles

Found 2 Documents
Search

Low-resolution facial emotion recognition on low-cost devices Dwisnanto Putro, Muhamad; Litouw, Jane; Poekoel, Vecky Canisius
IAES International Journal of Artificial Intelligence (IJ-AI) Vol 13, No 2: June 2024
Publisher : Institute of Advanced Engineering and Science

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.11591/ijai.v13.i2.pp2201-2211

Abstract

The low-resolution input image is a crucial challenge for applying facial emotion recognition in real-world scenarios. The critical problem is that valuable object features are relatively lost in the extraction process due to their small size. On the other hand, this vision system is required by a machine to run smoothly on lowcost devices. Facial emotion recognition using a lightweight feature extractor is proposed in this study to effectively capture crucial facial components in a lowresolution image. To compromise the running speed, this work offers an efficient feature convolution to discriminate specific facial features. In addition, the system is embedded with an attentive module to capture important features and correlate them. Our model performance is evaluated on low-resolution public datasets achieving the accuracy of 97.34%, 81.10%, and 80.12% on Karolinska directed emotional faces (KDEF), real-world affective faces database (RFDB), and facial expression recognition 2013 plus (FER2013Plus), respectively. The practical application demands that the deep learning model can operate fast on inexpensive devices. Consequently, the model achieved a speed of 290 frames per second (FPS) on a central processing unit (CPU) device.
Arsitektur Convolutional Neural Network Ringan Untuk Pengenalan Ekspresi Wajah: Lightweight Convolutional Neural Network Architecture For Facial Expression Recognition Robot, Reynold; Karenia Lolowang, Miesje; Dwisnanto Putro, Muhamad
Jurnal Teknik Elektro dan Komputer Vol. 13 No. 01 (2024): Jurnal Teknik Elektro dan Komputer
Publisher : Universitas Sam Ratulangi

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.35793/jtek.v13i01.56481

Abstract

Abstract — Facial expression recognition presents the challenge of applicability to real scenarios which demands to involve lightweight algorithms to operate at real-time speed. The proposed work offers a lightweight convolutional neural network architecture to accurately predict facial expressions. It considers the use of parameters and computations that are not massive. thus applying lightweight convolution operations. A channel compression technique at the center is applied to reduce parameter usage and redundant operations. Testing the accuracy results was conducted on the KDEF dataset. This dataset is a knowledge source that provides information on seven facial expressions with varying poses. Augmentation techniques were used to increase the variety and training configuration strategies were applied to optimize the network training work. As a result, the proposed model achieves an accuracy of 0.9732 which outperforms competing CNN architectures. Besides, it also produces a lightweight parameter of 2.8 M which can encourage the model to operate fast on Jetson Nano devices. The proposed architecture produces an effective deep learning model for predicting facial expressions from a chunk of input image without compromising its efficiency. Key words— convolutional neural network; Facial expressions; lightweight architecture; model efficiency.   Abstrak — Pengenalan ekspresi wajah menghadirkan tantangan penerapan pada skenario nyata yang menuntut untuk melibatkan algoritma yang ringan agar dapat beroperasi dengan kecepatan waktu nyata. Pekerjaan yang diusulkan menawarkan sebuah arsitektur convolutional neural network yang ringan untuk memprediksi ekspresi wajah secara akurat. Ini mempertimbangan penggunaan parameter dan komputasi yang tidak massif sehingga menerapkan operasi konvolusi yang ringan. Teknik pemampatan kanal pada bagian tengah diterapkan untuk mengurangi penggunaan parameter dan operasi berlebihan. Pengujian hasil akurasi diadakan pada dataset KDEF. Dataset ini merupakan sumber pengetahuan yang menyediakan informasi tujuh ekspresi wajah dengan pose yang bervariasi. Teknik augmentasi digunakan untuk memperbanyak variasi dan strategi konfigurasi pelatihan diterapkan untuk mengoptimalkan pekerjaan pelatihan jaringan. Sebagai hasil, model yang diusulkan meraih akurasi sebesar 0.9732 yang mengungguli arsitektur CNN pesaing. Disamping itu, ini juga menghasilkan parameter yang ringan sebesar 2.8 M yang dapat mendorong model untuk beroperasi cepat pada perangkat Jetson Nano. Arsitektur yang diusulkan menghasilkan model pembelajaran mendalam yang efektif untuk memprediksi ekspresi wajah dari sebuah potongan masukan citra tanpa mengabaikan efisiensinya. Kata kunci — Arsitektur ringan; convolutional neural network; efisiensi model ; ekspresi wajah