Abstract
Blur gradient is acknowledged as an important cue for relative distance. A number of studies examined detection and discrimination of blur, but no study to date has directly examined human performance of gradient detection per se. In this study we examined gradient detection by using filtered random patterns. Blur gradient was generated by applying low-pass filters with varied cut-off frequencies to random dot patterns. The random dot pattern was comprised of 10 x 10 min cells, and extended 5 x 5 deg. The cut-off frequency at the center of the pattern was either 1.6, 3.2, 6.4, or 12.8 c/deg, and it was gradually increased or decreased along the horizontal axis to generate a blur gradient. Observers were asked to detect the direction of the gradient either to the right or to the left. It was found that the visual system is very sensitive to blur gradient. The lowest threshold of 0.016 octave/deg was found for the center frequency of 3.2 and 6.4 c/deg. The threshold increased at both 1.6 and 12.8 c/deg. These results were compared to the blur-discrimination performance obtained with two 5 x 5 deg homogeneously blurred patterns placed side by side with a 1 deg gap in between. The same four cut-off as the main experiment was used for the standard stimulus. It was found that the gradient threshold is almost comparable to that of discrimination if the gradient detection is based on the comparison between two extreme positions within a pattern, although it is not likely that such extreme positions are used for gradient detection. In addition, the discrimination threshold increased monotonically as cut-off was increased whereas the gradient threshold followed a U function. These results, thus, suggest that gradient detection is based on a mechanism separate from that for blur detection.
Meeting abstract presented at VSS 2015