An improved Image Description Method Using Recurrent Neural Network with Gated Recurrent Unit

Image description is a comprehensive task combining natural language processing and computer vision. This paper proposes a model improved by GRU for the lack of long-term memory due to use of ordinary RNN in m-RNN. The new model can predict the words at the next moment based on the information at the previous moment, and generates a complete image description sentence. Moreover, using the MSCOCO2014 dataset, our model was trained and tested, and the indicators such as BLEU, METEOR and CIDEr are used as evaluation criteria. The experimental comparison with the original model and other models is carried out to further verify the validity of our model.