Multilayered Analysis of Newspaper Structure and Design

Abstract : Understanding newspaper structure and design remains a challenging task due to the complex composition of pages with many visual and textual elements. Current approaches have focused on simple design types and analysed only broad classes for the components in a page. In this paper, we propose an approach to obtain a comprehensive understanding of a newspaper page through a multi-layered analysis of structure and design. Taking images of newspaper front pages as input, our approach uses a combination of computer vision techniques to segment newspapers with complex layouts into meaningful blocks of varying degrees of granularity, and convolutional neural network (CNN) to classify each block. The final output presents a visualization of the various layers of design elements present in the newspaper. Compared to previous approaches, our method introduces a much larger set of design-related labels (23 labels against less than 10 before) resulting in a very fine description of the pages, with high accuracy (83%). As a whole, this automated analysis would have potential applications such as cross-medium content adaptation, digital archiving, and UX design.
Complete list of metadatas

Cited literature [13 references]  Display  Hide  Download

https://hal.inria.fr/hal-02177784
Contributor : Hui-Yin Wu <>
Submitted on : Tuesday, July 9, 2019 - 1:25:17 PM
Last modification on : Monday, July 22, 2019 - 2:42:04 PM

File

RR-9281.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-02177784, version 1

Collections

Citation

Hui-Yin Wu, Pierre Kornprobst. Multilayered Analysis of Newspaper Structure and Design. [Research Report] RR-9281, UCA, Inria. 2019. ⟨hal-02177784⟩

Share

Metrics

Record views

39

Files downloads

365