Image Summarization Using Unsupervised Learning

Recent trends show that a large amount of image data generated rapidly. Image Collection Summarization used to provide an automated collection of condensed representations for the collection of images. Most existing works on image summarization follow the supervised method. But the absence of appropriate ground truth data makes it impossible to solve the problem with traditional supervised methods. This paper proposes two unsupervised methods for extractive image summarization. The first method named Image Summarization using Clustering (ISUC) uses OneClassSVM for clustering followed by outlier detection to extract representative images. The second method named Image Summarization using Auto encoder (ISUA) uses auto-encoder to generate a representative score to each image in the collection and then the summary is created with images having diverse scores. Extensive experiments carried out on Div400 dataset shows the effectiveness of proposed methods over supervised methods and ISUA outperforms ISUC.