Figure 5.

The layout prototypes used in conjunction with CT to encode the spatial structure of scenes (see section 7.1). There are two different sets of such prototypes. One set of 10 prototypes is used for encoding the scene when the top ROI is fixated; the other set of 10 prototypes is used when the bottom ROI is fixated. For each situation (scene + fixation), the scene structure was thus represented by a 10-dimensional vector of similarities between the layout of the scene's ROIs and the 10 layout prototypes.