Image-based technique to select visually salient pages in large documents

The presentation of a book or a digital document is an important factor when designing the interface of a digital catalogue or an online bookstore. One way to attract readers and customers is to show a thumbnail or a preview of the book cover, a feature that can be extended to include a selection of the book's most appealing pages as well. If the number of documents handled by the system is large then the need to automatise the page selection task arises. This paper details an efficient and easily implementable image-based technique to choose a given number of pages to appear as thumbnails on a document list interface. The algorithm is based on a combination of rudimentary layout analysis and colour saliency computation in order to extract pages with attractive features such as large, colourful images and text elements. An arbitrary amount of spread can be injected in the selection process through an iterative filter function. A preliminary user study of the algorithm shows that its results are satisfying and provides insight as to how the technique can be improved.