Toward a Systematic Assessment of Map Reproduction Success Using OCR

Mulaw, Yohannes Abrha; Koukouraki, Eftychia; Kray, Christian

Abstract

Visualizations play a key role in presenting scientific results but surprisingly, research on assessing the reproducibility of scientific visualizations is scarce. In the absence of standardized criteria and metrics for evaluating reproduced visualizations, the research community has to rely mostly on subjective visual assessment. This issue becomes even more critical in the case of more complex visualizations such as maps, which frequently incorporate textual annotations, labels, and legends to provide context and assist interpretation. If the visual elements on the map change or disappear, the meaning and functionality of the map can be affected in several ways, from being difficult to use to providing misleading information. To set a starting point for this discussion, we explore the quantification of map differences focusing on textual elements with the use of optical character recognition (OCR). Aiming to facilitate more systematic reproducibility assessments, we developed an interactive web application that automatically extracts and parses textual elements from map images and computes similarity scores based on content and geometry.