Deep Learning Hybrid Technique for Generation of Image Caption

Image captioning is a fascinating and demanding work with applications in many different fields, including image retrieval, organizing and finding user-interested images, etc. It has enormous potential to replace the tedious process of creating captions for photos, and it works particularly well with massive amounts of picture data. Deep neural network-based techniques have recently shown significant success in the domains of language synthesis, machine translation, and computer vision. In this research, we present a paradigm based on encoder-decoders that may produce grammatically acceptable image captions. This model uses LSTM as the decoder and VGG16 Hybrid Places 1365 as the encoder. Model is evaluated using all standard metrics such as BLEU. Experimental results indicate that the proposed model obtained a BLEU-1 score 0.603774, BLEU-2 score 0.388514, BLUE-3 score 0.244706 on Flickr8k dataset. By comparison with the state-of-the-art techniques, the suggested strategy produced a noteworthy performance. We also present the outcomes of caption creation from real sample photographs, which supports the validity of the suggested method, in order to assess the model's effectiveness even further.