Normalization and uncertainty effects in three objective tasks using first-order and second-order textures

S. Sabina Wolfson, Norma Graham, & Oleg Slinin

Department of Psychology, Columbia University, New York, NY, 10027, USA

Vision Sciences Society (2004), abstract #E56, page 134


A contrast-gain control (normalization) has been demonstrated by having observers subjectively rate the perceived texture segregation of element-arrangement textures from constant-difference series of such textures (Graham and Sutter, 2000). Here the perception of these textures is investigated using three objective tasks: (A) Region Segregation -- Identifying the orientation of a rectangular region of an element-arrangement texture embedded in a background of another; (B) Texture Identification -- Identifying the orientation of the stripes in a striped element-arrangement texture; (C) Texture Detection (in Certain and Uncertain conditions) -- Detecting the presence of a striped element-arrangement texture.

Each texture was composed of two types of element. The two types differed in contrast but were identical in spatial characteristics (either a Gabor patch or a Gaussian blob). The two types were arranged either to form a striped- or a checkerboard-arrangement texture.

Observers' did much less well on the region segregation task than on the other two tasks. Nonetheless, the signature of normalization was seen in all three tasks.

The detection and identification results using Gabor-element textures demonstrated two properties of second-order (complex) channels previously shown for first-order channels. (1) The relationship between identification and detection can be explained by the existence of channels that are independent of one another. In the case of these second-order patterns, the independent sets of channels are second-order (complex) channels sensitive to the vertical and to the horizontal striped- element arrangements respectively. (2) The amount by which observers do better when certain than uncertain can be explained by assuming that on each trial observers can ignore any channels that they know will not give useful information (sometimes referred to as "excluding distracters").


Supported by NIH grant and EY08459.