Deep Learning to Automate Infographic Design

How artificial intelligence can improve your visual design skills.

Photo by Markus Spiske on Unsplash

Infographics are visual representations of datasets aimed to help people read and understand data more effectively. Creating effective infographics is not easy. Designers need to consider perceptual effectiveness, aesthetics, memorability, engagement, and different kinds of resource limitations. To help people understand and become more engaged with infographics, there are many different elements and styles that can be used in creating infographics — for example, colors, shapes, icons, texts, images, etc. Although there are many powerful design tools, the process of designing effective infographics is very demanding due to the lack of diversity. Tools like Microsoft PowerPoint and Adobe Illustrator are able to generate infographics automatically, but they provide only a limited variety of templates.

An automated approach to extract an extensible timeline template from a bitmap image. a) Original bitmap image; b) Content understanding including global and local information of the timeline; c) Extensible template contains editable elements and their semantic roles; d) New timeline (with mock-up colors) automatically generated with updated data.

In this work, Chen and his colleagues contribute a tool able to extract extensible templates from bitmap infographics, specifically timeline infographics. A timeline presents interval event data that can be described from representation, scale, and layout which also called global information in this paper. Relatively, the local information refers to the category, location, and the pixel-wise mask of each individual element.

To begin with, the researcher set two main goals to achieve: 1) the machine should be able to parse the content of images; 2) the machine should be able to construct extensible templates automatically. To achieve the goals, the researchers use a two-step approach: Deconstruction and Reconstruction. In the Deconstruction phase, the machine parses structural information from the bitmap timeline infographics. The researchers use existing architectures like Convolutional Neural Networks and ResNeXtIn to interpret the global and the local information of bitmap infographics.

Categories of elements in a timeline infographic. Information such as the event mark, annotation mark, and main body can be reused, annotation text, icons, event text can be replaced by updated data.

In the Reconstruction phase, the machine takes structural information obtained from the first phase as the input and then reconstructs extensible templates. The researchers use existing techniques and algorithms such as Non-Maximum Merging, Redundancy Recovery, GrabCut, and API to extract information for reuse and create extensible templates. The final approach is able to extract templates from timeline infographics and reuse graphical elements with updated data.

In the future, the research team wants to generalize the method and explore a larger variety of infographics.

This blog post is based on the following article:

• Zhutian Chen et al. Towards Automated Infographic Design: Deep Learning-based Auto-Extraction of Extensible Timeline. IEEE Transactions on Visualization and Computer Graphics, 2019.