Speeding up scientific imaging workflows: Design of automated image annotation tool

Low cost digital cameras have transformed the process of collecting data. Researchers can now generate massive datasets and analyze the images later, either manually or with the assistance of software tools for processing and annotating images. However, it remains time-consuming and expensive to develop custom software analysis tools for a specific research problem or domain - and often these custom tools cannot scale to larger datasets or adapt to new research questions. Existing image analysis tools also work best with well-defined research projects, where the researchers know what information to extract from each image. Yet, for new projects, it is especially difficult to build useful software, where researchers have not yet determined what information is “interesting” within the images. One way to increase the efficiency of the research is to improve the workflow of this exploration process. This paper presents one approach for improving exploratory image analysis workflows using point-based image annotation. We describe a landmark labeling system that (1) assists researchers in identifying interesting features and annotating images, (2) evolves over time to automate the annotation process, and (3) can be readily scaled and adapted to explore new problems and new domains. We describe both a proof-of-concept system in current use, and ongoing work to develop a generalizable software tool to support fully automated image annotation, with the ultimate goal of allowing researchers to explore data faster and significantly reduces the mean time to science.