A two-phase study was conducted which related the confusions I among dot matrix characters to the two-dimensional spatial frequency similarity of these characters.

During the first phase of the study, subjects were shown single alphanumeric characters from four different dot matrix fonts. In addition to the font variable, the size of the character was varied. All common matrix sizes, 5 X 7, 7 X 9, and 9 X 11, were used. The design of the study allowed the effects of matrix size (number of dots) and character size (angular subtense) to be separated in the analysis. Data from this phase of the research were analyzed in terms of both correctness and character confusion frequencies. The ANOVA of the number of correct character recognitions provided interesting interaction effects among font and matrix/character size. These results are discussed in terms of display design considerations.

The second phase of the study consisted of digitizing and analyzing all characters from two of the fonts used in the first phase. The fonts chosen represent the most and least confusable of the four, based on the performance data obtained. These characters were scanned photometrically using a computer-controlled X-Y stage. The resultant digitized arrays were subjected to a 512 X 512 point fast Fourier transform (FFT). The Fourier coefficients were correlated for all possible character pairs within each font-matrix/character size cell. These correlations provided an objective similarity measure among characters based upon their 2-D spatial frequency spectra.

In addition to the spatial frequency similarity measure, a simple digital Phi coefficient was calculated for each character pair. This coefficient is simply a nonparametric correlation coefficient between two digital arrays.

The final analysis performed in this study was the correlation of observed performance (confusions) with objective similarity measures (2-D spectra and Phi coefficients). A strong relationship between objective and subjective confusability would be a very useful design aid for display manufacturers. The obtained correlational relationships are discussed in terms of their utility for design and their implications for visual system models based on spatial frequency analysis.