Automated image captioning
Adela Pop Puscasiu, Alexandra Fanca, Dan-Ioan Gota, Honoriu Vălean · 2020
The research of comprehension of machine learning algorithms and procedures suitable for both image processing and natural language processing. The accommodation with existing packages used in implementing machine learning algorithms. The implementation of an algorithm that takes an image and describes it in comprehensive sentences. After the analyzed requirements, a bibliographic study was conducted for familiarization with the domain and possible models. Next, a study of the available technologies which would help to build the application was made, and then the design, implementation, and validation of the solution began. This paper presents a composite model, consisting of a deep convolutional neural network for feature extraction that makes use of transfer learning, and a recurrent neural network for building the descriptions. Keras with TensorFlow backend is used for implementation. A trained model that describes images using natural language was obtained.