We have studied some of the design trade-offs governing visual representati
ons based on spatially invariant conjunctive feature detectors, with an emp
hasis on the susceptibility of such systems to false-positive recognition e
rrors-Malsburg's classical binding problem. We begin by deriving an analyti
cal model that makes explicit how recognition performance is affected by th
e number of objects that must be distinguished, the number of features incl
uded in the representation, the complexity of individual objects, and the c
lutter load, that is, the amount of visual material in the field of view in
which multiple objects must be simultaneously recognized, independent of p
ose, and without explicit segmentation. Using the domain of text to model o
bject recognition in cluttered scenes, we show that with corrections for th
e nonuniform probability and nonindependence of text features, the analytic
al model achieves good fits to measured recognition rates in simulations in
volving a wide range of clutter loads, word sizes, and feature counts. We t
hen introduce a greedy algorithm for feature learning, derived from the ana
lytical model, which grows a representation by choosing those conjunctive f
eatures that are most likely to distinguish objects from the cluttered back
grounds in which they are embedded. We show that the representations produc
ed by this algorithm are compact, decorrelated, and heavily weighted toward
features of low conjunctive order. Our results provide a more quantitative
basis for understanding when spatially invariant conjunctive features can
support unambiguous perception in multiobject scenes, and lead to several i
nsights regarding the properties of visual representations optimized for sp
ecific recognition tasks.