Using human experts' gaze data to evaluate image processing algorithms

Understanding the capabilities of the human visual system with respect to image understanding, in order to inform image processing, remains a challenge. Visual attention deployment strategies of experts can serve as an objective measure to help us understand their learned perceptual and conceptual processes. Understanding these processes will inform and direct image the selection and use of image processing algorithms, such as the dermatological images used in our study. The goal of our research is to extract and utilize the tacit knowledge of domain experts towards building a pipeline of image processing algorithms that could closely parallel the underlying cognitive processes. In this paper we use medical experts' eye movement data, primarily fixations, as a metric to evaluate the correlation of perceptually-relevant regions with individual clusters identified through k-means clustering. This test case demonstrates the potential of this approach to determine whether a particular image processing algorithm will be useful in identifying image regions with high visual interest and whether it could be a component of a processing pipeline.