A Comprehensive Investigation on Image Caption Generation using Deep Neural Networks

• Autor(a) principal: Naik, Haraprasad
• Data de Publicação: 2022
• Outros Autores: Tripathy, Saswati, Laxmi Priya, Sahu, Prajna Paramita
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: INFOCOMP: Jornal de Ciência da Computação
• Texto Completo: https://infocomp.dcc.ufla.br/index.php/infocomp/article/view/1495

Caption Generation from an Image is a task that has close relationship with Object Detectionand Natural Language Processing. However, The Object Detection has been evolving tremendouslysince a decade. In the Traditional approach of Object Detection there is an involvement of three stepprocess, which includes (1) region selection (2) Feature Extraction and (3) Classification. But in currentdays research trends Neural Networks are used to overcome the hand-crafted feature extraction alongwith the application of classification algorithm through various algorithm such as SVM, AdaBoost andDeformable Part Model (DPM). Similarly, to generate a Caption we can take the help of Neural Networkspecifically Recurrent Neural Network (RNN). In most of the Caption Generator a variation of RNNis used, that is Long Short-Term Memory (LSTM). Many researchers adopts either sampling or BeamSearch to generate a valid sentence/caption. In this paper our aim is to capture the fundamental ideabehind the presently available Image Caption Generator and compare their architectural design withtheir performance.