Identification of filtered lettersin filtered noise

(Presented at VSS,2002)

Lauren F. V. Scharff, Albert J.Ahumada

Stephen F. Austin State University, NASAAmes Research Center

Several researchers have investigated theeffects of filtering letters and/or added noise on letteridentification (e.g. Legge, Pelli, Rubin & Schleske, 1985(letters filtered but no noise added); Parish & Sperling, 1991(noise filter matched letter filter); Solomon & Pelli, 1994(noise filtered but letters unfiltered)). This study compares thecombined effects of high and low pass filtering on both letters andnoise. Letters (Arial font, 24 pt) that were unfiltered, high- orlow-band pass filtered (sharp cut off frequency of approximately 4cycles per capital letter height) were presented in unfiltered, high-or low-band pass filtered noise (9 combinations). Three participantsmade 26-alternative, forced-choice responses for each of the 9conditions (counterbalanced, blocked presentation) for both upper andlower case letters. For each condition, three replications of 30trial threshold estimations were run, in which letter contrast wasadjusted by a Quest algorithm. Averaged thresholds showed that for agiven noise, unfiltered letters (the sum of the high- and low-passletters) led to better recognition than either component filteredletter alone. However estimates for summation indexes in thedifferent noises were strikingly different. In the wide band noise,summation was between linear and Euclidean. In high-pass noise, thesummation was quite linear. But, for low-pass noise, there was nosummation. Adding the low-pass letter component did not improveperformance; high-pass letters in low-pass noise led to betterperformance than unfiltered letters in low-pass noise. The resultsqualitatively support a moveable single filter model where the filterband was moved most in the low-pass noise and least in the high-passnoise.


Supported by NASA RTOP 548-54-12.