Abstract
An image from Beck (1966) has been influential in textbook presentations of texture segmentation and has driven development of theories of texture segmentation and computational models of grouping. The image consists of left, middle, and right regions, where each region contains 36 copies of a backwards letter L, T, or slanted T, respectively, in the same relative locations. The reader is told that people indicate weaker segmentation between the L and T regions than between the T and slanted T regions, thereby indicating that texture segmentation is not based on shape similarity. More generally, the Beck (1966) study investigated the roles of orientation and similarity on texture segmentation by varying letter type and orientation. In Experiment 1 the task was to indicate which side of an image had the “more natural” break. For Experiment 2, participants viewed two of the regions and rated their similarity using a 10-point scale. Beck concluded that reported segmentation was based on common orientation and not similarity. Due to the seminal role of Experiment 1 and its small sample size (n=16), we initially planned to replicate Experiment 1 only using a larger sample size (n=101). The experiment only partially replicated (r=0.6). Two issues made the experiment difficult to replicate: the example stimuli and textual description are incompatible, and it seems that spacing between elements was not properly controlled in the original study. Additionally, a reanalysis finds that the data in Beck (1966) actually does suggest a role of similarity in texture segmentation: more similar regions tend to be the chosen segmentation (r=-0.7). Since this conclusion is counterintuitive, we also successfully replicated Experiment 2 (n=23, r=0.9). The reported segmentations in our Experiment 1 were partly explained (r=0.4) by the similarity ratings in our Experiment 2: more similar regions were less likely to be the chosen segmentation.