Image Captioning Using CNN-LSTM

We regularly encounter a large number of images from various fields such as news articles, diagrams, literature, and the Internet. Even though several images do not have a description, humans have the capability to understand them without their captions but this is not the case with machines. As long as machines do not behave, think, and talk like humans, natural language descriptions will be challenging to solve. Although image captioning is a challenging undertaking, numerous researchers have made substantial advancements. Image captioning is the task of automatically generating a textual description of an image. This technology has significant potential for a range of applications, including assistive technologies for visually impaired individuals, content-based image retrieval, and personalized recommendation systems. In this work, CNN-LSTM model is used to generate captions for images by processing its features. This paper mainly describes an image captioning method using a deep learning approach, a combination of LSTM and Xception model, which shows representative work of the model. Xception is used to extract the high-level features of the images. The extracted features alongside the image captions are fed as input to the LSTM which generates the most appropriate caption for the given image. In this work, the model is evaluated using BLEU score, a standard metric that is used for evaluating machine-translated text. This will summarize the images without any human intervention.