The role of features in letter detection and identification

It is widely believed that pattern detection is mediated by many independent mechanisms, "feature detectors". This is the key assumption underlying probability summation, for which there is ample empirical support. However, it seemed unlikely to us that such an inefficient scheme would mediate our ability to identify important familiar patterns, like letters. In fact, Pelli, Burns, Farell, and Moore (in press) report a fixed ratio of thresholds for identification and detection, across a wide range of typefaces (including bookman, bold bookman, and script K script U sign̈ script N ℐ script T sign ℒ ℰ ℛ). We will show that this fixed ratio is consistent with the idea that several (between 2 and 5) features are detected at the identification threshold.