Object-aware text placement for banner images

Abstract

In the field of graphic design, generating layout consisting of visual and textual elements is a challenging task. We propose a method to estimate a text position from the background image. Text elements are located so that the text element does not hide the main content of an image. Based on this observation, we detect objects in a background image to estimate text position. We also developed an interactive interface where users can manually scale and crop the background image. Our system changes the text position according to the user input in real-time. We compare our method to a saliency-based baseline. The results suggest the effectiveness of our method. Our system will be demonstrated at the interactive session.