Modelling Visual Salience in Real-Time

As part of my Ph.D. I developed and algorithm that models the visual salience of the objects rendered in each scene as a user moves through a simulated 3-D environment. The basic assumptions underpinning this algorithm are: the closer an object is to the center of the scene the greater its salience and the larger an object is the greater its salience.

This algorithm extends the false colouring technique (Kuffner and Latombe 1999; Peters and O'Sullivan 2002). In the false colouring technique each object in the simulation is assigned a unique RGB value (it's false colour). Then in parallel with the normal rendering of each scene an off screen rendering using the false colours is computed. This offscreen rendering is then scanned and a list of the colours in the scene is created. This list identifies all the objects in the scene. The image below left shows the normal rendering of a scene, below right illustrates the false colour redering of the scene. A journal article describing this visual salience algorithm was published in Artifical Intelligence Review Vol. 21(3):Visual Salience and Reference Resolution in Simulated 3-D Environments.