Feature synergy enhances detection but not recognition of shape from texture cues

Abstract

Texture regions that differ from their surroundings in more than one local feature are more easily detected. Recent findings show that a low-level summary statistic, net contrast energy, predicts this double-cue advantage, suggesting early-stage integration during image analysis. We investigated whether this advantage also applies to more complex, texture-defined shape discrimination beyond figure-ground segregation. Using both a figure detection task and a more demanding shape identification task, we calibrated sensitivity to fixed baseline levels with single-cue targets defined by orientation or spatial frequency contrast. We then measured performance for double-cue targets at these baselines. Contrary to earlier results reported for simpler shape discriminations, we found a reduced double-cue advantage in the shape identification task. Specifically, double-cue sensitivity was notably lower than the algebraic sum of the single-cue sensitivities, a level achieved consistently in the detection task. Control tests with high feature contrast showed perfect detection performance for both single and combined cues. However, shape identification saturated at levels between 83%–90% accuracy, while gray-shaded figures yielded perfect performance, suggesting that unique shape representations could not be built from single or combined texture cues. These findings suggest that texture cue summation enhances texture segregation and segmentation but does not improve higher-level recognition of 2D texture shapes.