Item

Abstract

Countless climate images are in circulation on the internet, such as burning globes, polar bears and photos of global climate impacts. These images are networked and generate a specific view of climate change. Our case study engages in an intercultural image comparison based on Google Image queries. As an interdisciplinary team of experts drawn from art his- tory, media studies, interface design and computer graphics, our goal was to use a combination of qualitative and quantitative image analyses to explore the predominant discourses of digi- tised visual climate communication on the web. To this end, we automated the analysis of dif- ferent formal features of climate images (such as colour values, density and composition) with the aid of computer-driven methods (such as computer vision and machine learning) to build a corpus of thousands of images. Our focus was on image similarities, a concept shared by both image theory and computer analysis. In this chapter, we elucidate the outcome of our research on a conceptual and technical basis. The core issue addressed here is the manner in which art- historical methods (such as iconography and the concept of visual framing) are transformed when using computer-generated methods of computer vision and machine learning to anal- yse image similarities. This chapter focuses on our various insights while also reflecting on the general question of networked images on a methodological level. Ultimately, we were able to identify the promising potential but also the key limits of algorithmic image recognition and sorting when using machine learning to study images on the internet.