1. SIGNAL DETECTION IN FIXED PATTERN CHROMATIC NOISE 1 A. J. Ahumada, Jr., 2 W. K. Krebs 1 NASA Ames Research Center; 2 Naval Postgraduate School, Monterey, CA. PURPOSE The goal is to develop image discrimination models that can predict the detectability of targets in color images. Previous work with luminance targets in complex imagery has shown that target detectability is strongly influenced by background contrast, but little data is available to calibrate and test color image discrimination models, which require the same background with and without the target. We measure the masking of simple targets by fixed pattern chromatic noise. These measurements simplify the testing and calibration of the contrast masking components of color image discrimination models. METHODS Observers. The 10 observers were students who had passed standard tests for acuity and color vision. Stimuli. The color CRT was viewed from 1 m, with 37.6 pixels per degree of visual angle in the vertical direction and 52.7 pixels per degree horizontally. The screen background (1024 x 512) was yellow with a luminance of 18.4 cd/m 2. The fixed pattern noise images (masks) were 138 by 100 pixels, generated as 69 x 50 images and then pixel replicated. The luminance masks were the background yellow. The red/green masks were isoluminant, as defined by a Minolta CS-100. The noise sample forming the red/green mask was different from that of the luminance mask, but the sum mask was the sum of those two images. A similar low contrast set of three masks was formed using two new noise samples at half the contrast. For the high contrast set, the peak contrast was 0.5. Two targets were used. They were red/green uniform squares (14 x 10 pixels). The luminance target was the background yellow. For the red/green target, the green gun level decreased as the red increased, keeping the luminance constant. The color of the red/green target varied from red to the background yellow as the target contrast (defined as the red signal contrast) was lowered. Procedure. Thresholds were estimated using a 2IFC QUEST procedure for blocks of 30 trials. The signal and no-signal intervals were each 0.5 sec in duration and separated by an interval of 0.5 sec. A feedback tone indicated an incorrect response. Horizontal and vertical lines outside the masking image region indicated the position of the possible target during the trial. Each observer ran one practice day and three test days. On each day the observer did a warm-up session of 5 blocks of conditions chosen at random without replacement from the 14 conditions, and then did one block for each masking condition in a random order. For each observer and each masking condition, the medians of the 3 daily threshold contrasts were found, then the masking effect represented in decibels (20 log10(masked threshold / unmasked threshold)). BEHAVIORAL RESULTS MODEL PREDICTIONS Predictions are shown for an image discrimination model with contrast masking computed independently in the luminance and red- green channels. Only the prediction of no masking of the luminance target by the red/green noise (YR) is outside the observer data confidence interval. The average noise contrast level effect is predicted to be 3.1 dB, also in its confidence interval. CONCLUSIONS Luminance noise did not mask the red/green target. Red/green noise masking of the luminance target was small, but significant. The sum results show this latter effect could only be seen with masking imagery of very low luminance contrast. Thus, masking was fairly well predicted by a model with contrast masking within channels only. High Contrast Red/Green Noise (R) MASKS High Contrast Luminance Noise (Y) Summed High Contrast Noises (S) Low Contrast Red/Green Noise (r) Low Contrast Luminance Noise (y) Summed Low Contrast Noises (s) Confidence intervals (95%) for these measurements are plus or minus 3.0 dB. The results show small but significant masking of the luminance target by the red/green noise (4.1  3.0 dB). The luminance noise alone did not mask the red/green target, and the increased masking by the sum mask over that of the red/green mask alone is not significant (4.1  5.0 dB). The average difference for the high and low contrast maskers was 4.0  1.4 dB. The luminance target predictions are simple: luminance targets do not affect the red/green channel and are unaffected by red/green noise. The luminance noise slightly enhances the red/green signal because the red/green channel is formed by dividing a linear red/green signal by the local luminance, which contributes more when the luminance is low than it loses when the luminance is high. TARGETS on the UNIFORM BACKGROUND Luminance (Y)Red/Green (R)