Facial Expression Recognition and Image Description Generation in Vietnamese

This paper discusses a facial expression recognition model and description generation to build descriptive sentences for images expressions of people in images. Our study shows that YOLOv5 achieves better results than traditional CNN all emotions on the KDEF dataset. In particular, accuracies models emotion are 0.853 0.938, respectively. A generating descriptions based merged architecture is proposed using VGG16 with encoded over an LSTM model. also used recognize dominant colors objects correct color words generated if it necessary. If contains referring person, we person image. Finally, combine create describe visual content human Experimental Flickr8k dataset Vietnamese achieve BLEU-1, BLEU-2, BLEU-3, BLEU-4 scores 0.628; 0.425; 0.280; 0.174,