Knowledge-driven Gaze Control in the NIM Model
Joyca P. W. Lacroix, Jaap M. J. Murre, Eric O. Postma · eScholarship (California Digital Library) · 2006
In earlier work, we proposed a recognition memory model, the Natural Input Memory (NIM) model, that operates directly on digitized natural images.When presented with a natural image, the NIM model employs a biologically-informed perceptual preprocessing method that takes local samples (i.e., eye fixations) from the image and translates these into a similarityspace representation.Recognition is based on a matching of incoming and previously stored representations.In this paper, we investigate whether it is possible to extend the NIM model with a gaze control mechanism to select relevant eye-fixation locations based on scene-schema knowledge and episodic knowledge.We perform two experiments.In the first experiment, we test whether the similarity-space representations can be used to infer scene-schema knowledge of a specific category of natural stimuli, i.e., natural face images.In the second experiment, we examine how the model can use the scene-schema knowledge in combination with stored episodic knowledge to direct the gaze toward relevant spatial locations when performing a categorization task.Our results show that the spatial structure of face images can be inferred from the NIM model's similarity-space representations, i.e., scene-schema knowledge can be acquired for the category of face images.Moreover, our results show that extending the NIM model with a gaze control mechanism that combines scene-schema knowledge with stored episodic knowledge, enhances performance on a categorization task.