The Complexity of Perceptual Search Tasks

This research was motivated by the fol lowing question: What is the inherent computational d i f f i culty of the visual search experimental paradigm in psychology? This is an important issue since so many computational models have recently appeared that use visual search data as mot ivat ion. How can these results be properly used if the computational nature of the experiment itself is not understood? A computational def ini t ion of the visual search task is presented, and the bottom-up case is distinguished f rom the task-directed case. Then, a proof is given showing that the bottom-up case is NP-Complete in the size of the image, while the task-directed case has linear t ime complexity in the number of items in the display. The NP-Completeness of the bottom-up case is due solely to the inabil i ty to predict which pixels of a test image correspond to objects in a non-exponential manner. This provides the strongest possible evidence for the abandonment of purely bottom-up schemes that address the ful l generality of v is ion. It is thus necessary to sacrifice generality in order to re-shape the vision problem and to opt imize the resources dedicated to visual information processing so that a tractable problem is addressed.