Image Montage Summarization

Describing an image’s content automatically is a basic issue in artificial intelligence that links computer vision and processing of natural language. Recently, however, less attention has been given to extracting summaries from a set of associated pictures that can provide much better data. This paper presents an abstractive summary model with an Encoder-Decoder hierarchy that simultaneously sums up a gallery of pictures and matches phrases and pictures in summaries. The model is designed in order to enhance the probability of the destination identification sentence given the teaching picture. The precision of the model and the fluency of the language learned so only from image descriptions are demonstrated in experiments on various datasets. Our model is often quite precise and we check it in qualitative and quantitative terms. A recent study on neural summarization shows the power of the encoder-decoder model for picture and document overview. Experiments demonstrate that our model is better than neural abstraction and extraction techniques by producing better informative summaries of the collection of images.