Image Captioning Using Machine/Deep Learning

The System that automatically write descriptions or write captions for an image according to the contents of that image is known as Auto Image Captioning. It’s literacy problem of the machine that combines computer vision (for understanding image contents) with Natural processing (for textbook product) Auto picture captioning is a fairly new and fleetly adding academic content. Colourful new styles are being launched on a diurnal base in order to attain satisfying outcomes in this discipline. Still, to gain mortal- suchlike results, a lot of attention is still necessary. The thing of this exploration is to discover, in a methodical manner, what different and recent methodologies and models are employed for picture captioning using deep learning. What approach and new ideas are used to make these models better and more accurate. what’s further, ideas and strategy are more likely to give suitable result or outcome. To get so, we did a methodical literature analysis of recent exploration published between 2017 and 2019 from well- known datasource (Scopus, Web of Lores, IEEEXplore). We plant an aggregate of over 60+ primary studies that were related to this study. We found out that CNN understands the content in the picture and detect objects in an image, whereas LSTM/RNN is used to induce language. MS COCO, flicker 8k and flicker 30k, which are used in almost every studies, are the most extensively used datasets. The BLEU (1 – 4) standing rating matrix is the most extensively utilised in all projects or studies. The analysis also show that CNN with LSTM is more preferable than CNN with RNN. We found out that the Encoder Decoder and the attention medium are one of the most suitable and best ways for enforcing this model, and combining them can ameliorate outcomes to more accurate and good level. Experimenters or other people that want to work or contribute to auto image captioning will surely find this exploration very useful and can use it as a guidance.