A comprehensive literature review on image captioning methods and metrics based on deep learning technique

One of the trending areas of study in artificial intelligence is image captioning. Image captioning is a process of creating descriptive information for visual objects, image metadata, or entities present in an image. It extracts features from the image using the integration of computer vision and Natural Language Processing (NLP), uses this data to identify objects, actions, and the relationships among them, and creates image descriptions. It is not only an extremely important but also a very difficult task in computer vision research. A lot of work on image captioning methods that utilize a deep learning approach has been conducted. The goal of this article is to discover, evaluate, and summarize the works that examine deep learning applications in the context of image captioning systems. We found 548 papers using a systematic literature review (SLR) technique, of which 38 were identified as primary studies and so underwent in-depth analysis. This review's result demonstrates that LSTM, CNN, and RNN are mostly employ deep learning techniques for image captioning. Also, the most popular used datasets based on the selected primary studies are MS COCO Dataset, Flickr8k, and Flickr30k. These are standardized benchmark datasets being employed by researchers to compare their methods on common test-beds. The review also showed that the evaluation methods such as BLEU, CIDEr, SPICE, METEOR, and ROUGE-L are the most often employed ones according to the findings from this SMR study. Despite the considerable advancements achieved by deep learning approaches in this study domain, there is always a potential for improvement. Finally, the review provided future research for image captioning systems. We believe that this SLR will act as a reference for other scientists and an inspiration to gather the most recent data for their study evaluation.