Detection of image stretching

Abstract

Resizing images for different devices often involves changing the aspect ratio. A wide variety of approaches for resizing exist: sophisticated "content-aware" (or retargeting) approaches are built on the assumption that carefully chosen distortions are preferable to the naïve approach of uniformly stretching the image. However, there is little codified understanding of how distortions of the image, including uniform stretching or more complex warps introduced by retargeting, are perceived. In this paper, we describe experiments that explore the perception of image stretching, to establish the baseline for assessing more complex resizing methods, as well as to establish the methodology. In a series of experiments, we show that the perception of stretching is a complex phenomenon depending on a myriad of factors including the amount of distortion, the image content, the viewer's cultural background, and the observation time. We provide a methodology for creating images that avoid unfair cues to stretching and explore issues in using online worker communities for studies. We show that even small stretches can be detected in some cases. These findings have ramifications for the design and evaluation of image retargeting, and suggest that a more thorough study of distortion perception is necessary.