Clémence Bertrand Pilon^1,2 (), Martin Arguin^1,2,3,4; ¹Department of Psychology, Université de Montréal, Montreal, Quebec, Canada, ²Centre de recherche de l’Institut universitaire de gériatrie de Montréal, Montreal, Quebec, Canada, ³Centre interdisciplinaire de recherche sur le cerveau et l’apprentissage (CIRCA), Department of Psychology, Université de Montréal, Montreal, Quebec, Canada

The spatial frequencies (SF) optimal for word recognition are well established. Studies with other classes of stimuli however, demonstrate a rapid temporal evolution of the SFs most useful for visual recognition in a coarse-to fine order. The present study is the first to examine the time course of SF processing in a visual word recognition task. Word images were filtered according to four non-overlapping SF bandpass conditions (center frequencies of 1.2, 2.4, 4.8, and 9.6 cycles per degree). The stimulus presented on each trial was an additive combination of the target image (i.e. signal) and of a white noise field. The signal-to-noise ratio (SNR) varied randomly throughout the 200 ms exposure duration. Performance was maintained at 50% correct by adjusting the target contrast on every trial. Accuracy-based classification images (CI) of processing efficiency as a function of time were calculated to demonstrate the temporal progression of SF processing. These time-domain CIs show that the highest spatial frequency range dominates early processing, with a shift toward lower spatial frequencies later during stimulus exposure. This pattern interacted in complex ways with the temporal frequency content of signal-to-noise oscillations, as demonstrated by time-frequency CIs. A machine learning algorithm (support vector machine with leave-one-out cross validation) was exposed to the data patterns of individual participants with the task of deciding the SF band it corresponds to. The maximum classification accuracy of 90.6% correct (over 25% chance performance) was achieved when the classifier was exposed to the Fourier transformed time-frequency CIs. This level of accuracy is about twice that obtained from the classifier with raw or Fourier transformed time-domain CIs or raw time-frequency CIs. These findings suggest that the temporal progression of SF processing in visual word recognition is best understood if the time dimension is actually recast into its Fourier transform.

Acknowledgements: Supported by grants from the Fonds de Recherche Québec—Nature et Technologie and the Natural Sciences and Engineering Research Council of Canada (NSERC) to Martin Arguin and a research scholarship to Clémence Bertrand Pilon from the Conseil de recherches en sciences naturelles et en génie (CRSNG).