Abstract
The visual system can use multiple cues, including brightness, texture, color, and motion, to segment figure from background. We studied how two cues, texture and color, combine for visual shape recognition. Two experiments tested whether texture and color are processed independently for the identification of letters.
Methods. Observers identified 10x10 deg letters (Sloan font, 5 possible: DNRSZ) that were briefly displayed (250 ms) in the center of the screen. Letters were composed of oriented, filtered noise (1 cycle/deg peak SF, bandwidth 1 octave; orientation bandwidth 30 deg) in both foreground and background. We used "second-order" letters: The letter differed from background in dominant orientation (texture cue) and/or the hue-luminance association (color cue; dark-red/bright-green foreground and bright-red/dark-green background, or vice versa). The letter and background had identical mean luminance, hue (gray), and luminance contrast. Performance was determined as a function of second-order modulation contrast.
Results. Experiment 1: Letter-identification thresholds were determined for the texture cue alone, for the color cue alone, and for three two-cue conditions differing in relative chromatic and texture contrast. Compared to single-cue conditions, two-cue thresholds showed improvement consistent with pooling of independent cues. Experiment 2: Observers identified the letter defined by one cue (e.g., texture) while ignoring the other cue (e.g., color). The "ignored" cue signaled either the same letter (congruent trials), a different letter (incongruent trials), or no letter at all (baseline), randomly across trials. Compared to baseline, performance was better on congruent trials and worse on incongruent trials. Analysis of errors on incongruent trials revealed that observers often responded according to the irrelevant cue.
Conclusion. Human observers can integrate texture and color cues for improved shape recognition. While this improvement is consistent with independent processing of texture and color, observers cannot discount irrelevant color information when identifying shape based on texture, and vice versa.
Meeting abstract presented at VSS 2012