Leveraging Deep Learning Model for Image Caption Generation for Scenes Description
Vidyadevi G. Biradar, G. Mukund, Suyush Agarwal, Saurabh Kumar Singh, R Ujwal Bharadwaj
Abstract
The generation of image captions is an important task. with the recent advancements in deal learning paradigm, it is possible to develop models which are based on image feature extraction model and natural language processing to automate generation of image captions for a given image. Image captioning involves recognition of significant objects, features, and their associations with other objects in the given image and generate syntactically and semantically correct sentences. This paper presents detailed survey on recent methods used for automatically generating image captions. The contributions of the work include summarization on various techniques, recommendation of benchmark databases for training and testing image captioning models, implementation of model based on CNN and LSTM. The CNN model helps in the image feature extraction which is fed to the LSTM model to generate text in the natural language. The performance of the model is evaluated on Flickr 8k and model achieves a BLEU-I score of 0.755367.