Modelling cue integration in clutter
A complex, cluttered environment is replete with cues, and deciding which of them go together can be at least as great a challenge as properly integrating the ones that do. Although human observers are often imperfect in such settings, they usually outperform machine-perception algorithms on tasks such as demarcating and identifying objects seen against a cluttered background, or following a conversation in a noisy crowd. This failure to build machines capable of matching human performance may be seen as a sign that the field of perceptual studies has not yet found the right way in which to analyze such complex situations. Indeed, one of the largest challenges to making a successful analysis of perception and cue integration in such contexts comes at the outset in setting up an appropriate model. In this chapter we will take some modest strides towards building new cue integration models with the expressive power necessary to capture at least some of these complex settings.