The visual processing of text
The results of an investigation into the nature of the visual information obtained from pages of text and used in the visual processing of text during reading are reported. An initial investigation into the visual processing of text by applying a computational model of early vision (MIRAGE: Watt & Morgan, 1985; Watt, 1988) to pages of text (Computational Analysis 1) is shown to extract a range of features from a text image in the representation it delivers, which are organised across a range of spatial scales similar to those spanning human vision. The features the model extracts are capable of supporting a structured set of text processing tasks of the type required in reading. From the findings of this analysis, a series of psychophysical and computational studies are reported which exan-dne whether the type of information used in the human visual processing of text can be described by this modelled representation of information in text images. Using a novel technique to measure the 'visibility' of the information in text images, a second stage of investigation (Experiments 1-3) shows that information used to perform different text processing tasks of the type performed in reading is contained at different spatial scales of visual analysis. A second computational analysis of the information in text demonstrates how the spatial scale dependency of these text processing tasks can be accounted for by the model of early vision. In a third stage, two further experiments (Experiments 4-5) show how the pattern of text processing performance is determined by typographical parameters, and a third computational analysis of text demonstrates how changes in the pattern of text processing performance can be modelled by changes in the pattern of information represented by the model of vision. A fourth stage (Experiments 6-7 and Computational Analysis 4) examines the time-course of the visual processing of text. The experiments show how the duration required to reach a level of visual text processing performance varies as a function of typographical parameters, and comparison of these data with the model shows that this is consistent with a time-course of visual analysis based on a coarse-to-fine spatial scale of visual processing. A final experiment (Experiment 8) examines how reading performance varies with typographical parameters. It is shown how the pattern of reading performance and the pattern of visual text processing performance are related, and how the model of early vision might describe the visual processing of text in reading. The implications of these findings for theories of reading and theories of vision are finally discussed.