Automatic generation of description for images using recurrent neural network

Abstract

In this paper, a generative long short-term memory (LSTM) model for generating description of the image is implemented. Automatic generation of description that describes the content of a given image is a fundamental problem in artificial intelligence. This kind of work is achieved by connecting two different domains like computer vision and natural language processing. The solution proposed here makes use of deep learning. A deep learning framework known as Keras is used which uses TensorFlow for the backend process. TensorFlow is a framework used to do a series of operations in a chain. The general technique is to feed the features of an image to the model, which is capable of generating text of length less than or equal to a predefined caption length. The dataset Flickr30 K is used to train the model. The InceptionV3 is used to extract features of the images. BLEU metric is used to measure the accuracy of the description that is generated for that image using LSTM model.