EGITS: Expert Gaze-Guided Training System for Enhancing Image Interpretation Ability

Image interpretation requires the integration of explicit and tacit knowledge, yet existing interactive training systems struggle to convey expert tacit strategies to novices. We developed the expert gaze-guided image interpretation ability training system (EGITS), a human-computer interaction framework that combines cascade forest-based eye-movement classification (EMCCF) with cognitive strategy spatiotemporal similarity quantification (CSTM) to support gaze classification, expert-novice strategy comparison, and personalized feedback. Three experiments evaluated the system. CSTM showed superior robustness, outperforming a state-of-the-art benchmark by approximately 14%, and achieved significantly better performance in most gaze-similarity scenarios. In a three-phase online study, EGITS-trained novices showed a 122.5% increase in cognitive-strategy similarity to experts and significantly outperformed the control subgroup. Training was also associated with fewer fixations, longer average fixation durations, and shorter saccade amplitudes. These findings provide preliminary evidence that EGITS promotes expert-like visual-search strategies in FPS-based image-interpretation tasks and support the feasibility of eye-tracking-based visual-search training.