A replication and re-analysis of a classic texture segmentation study

A classic finding reported in Beck (1966a) is that observers tend to indicate a more natural texture break between a set of T's and tilted T's than between a set of T's and backward L's. This finding has played a prominent role in discussions about the properties of texture segmentation and in the development of computational theories of texture segmentation. Due to the small sample size of the original study, we replicated the original experiment with a larger sample. Regrettably, we discovered that the description in Beck (1966a) is insufficient to allow us to reproduce the stimuli. For our replication, we created stimuli that seem consistent with the spirit of the original study. The results of the replication study partly match Beck's original data ( $$ ), but conclusions indicated by these results deviate from some of Beck's key conclusions. We also further explored the influential claim derived from a second experiment in Beck (1966a) that texture segmentation was not related to the perceived similarity of elements. Contrary to this claim, a re-analysis of Beck's data indicates the counterintuitive conclusion that observers tend to indicate stronger texture segmentation between regions that are rated as being more similar. In a replication of the second experiment, this relation flipped so that the replication observers indicated stronger texture segmentation for regions with lower similarity ratings. Despite being influential, we conclude that there are substantial problems with the stimuli, analyses, and conclusions in Beck (1966a).