A deep learning-based system for automated image caption synthesis

Abstract Image captioning integrates Natural Language Processing(NLP) and computer vision to autonomously produce descriptions of texts for images. The proposed work introduces an image captioning system utilizing deep learning, incorporating Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and a visual attention mechanism (CNN+LSTM+Attention). The model emphasizes dynamic attention to pertinent image regions during caption generation, distinguishing it from conventional models by improving caption accuracy and producing more contextually relevant descriptions. The proposed model is evaluated on the Flickr8k dataset and compared with several state-of-the-art methods, including Transformer-based model, ResNet-101+SA-Bi-LSTM, and Guided Visual Attension (GVA). The proposed model demonstrated enhanced performance, achieving scores of BLEU-4 0.76, a METEOR 0.30, and a ROUGE-L score of 0.64, thereby surpassing both traditional models and more advanced Transformer-based approaches. The results evaluate the effectiveness of the proposed mechanism for producing the relevant and precise captions of the image. The proposed attention mechanism facilitates the concentration on the essential regions of an image. This paper presents the proposed model’s benefits in comparison to current State-of-the-art(SOTA) Techniques and highlights the future research directions in the inclusion of quantum-inspired and Transformer based models to augment the models capabilities.