Our previous research has established a number of similarities in the processing of textured stimuli by pigeons and humans (Cook, 1992a, 1992b, 1993b; Cook et al., 1996), which have suggested that the processes of early vision are analogously arranged in these dissimilar species. In trying to understand early vision in humans, and in particular the processes of perceptual grouping and texture segregation, several methods have been identified for isolating the separate contributions of the latter processes. The now classic example is Beck (1966). He established that the strength of global texture segregation reported by humans was not directly a function of the local similarity of the separate elements used to make the texture. In a similar vein, Wolfe (1992) has shown that visual features capable of supporting rapid visual search do not necessarily support strong texture segregation and vice versa. Such findings indicate that perceptual grouping processes result in the separate creation of new emergent attributes in the stimulus that are not a direct property of the individual elements per se, at least in humans.

The only previous attempt to examine this issue of the relation between perceptual grouping and element similarity in pigeons, however, suggested that this was not the case in birds (Blough & Franklin, 1985). These researchers tested pigeons with texture stimuli much like those tested here, but composed from letters of the alphabet. They compared their results with these textured stimuli with those previously obtained from a discrimination where the letters were tested as individual elements (Blough, 1982). Examining the pattern of stimulus confusions generated among the letters in each of these tasks, Blough & Franklin found no differences between the two display organizations. The correlation in reaction time (RT) performance with the different letter combinations over the two display types was high (r=.89), an outcome suggesting that the same set of features were employed to both cases.

One goal of the current experiment was to reexamine this specific issue, since Blough and Franklin=s results seemed at odds with the general pattern of texture results that we had established in pigeons. In the present experiment, the best source of information regarding this particular issue comes from comparing the birds= performance across the texture and geometric display types. To help with answering these types of questions, following the completion of transfer testing in Experiment 2 we expanded the number of shapes (the three transfer shapes plus one new shape), colors (the three transfer colors plus one new color), and objects (discussed later) tested in each daily session. Based on the 100 sessions of testing following this expansion, we then examined the correlation in choice performance among the texture and geometric display types for the 132 different shape and color element combinations tested with each display type. Shown in the figure is the bivariate correlation among these display types for each shape (left panel) and color (right panel) element combination.

Overall, the correlation was quite low for the different shape combinations (r=.25; for comparison the correlation in shape performance between feature and texture displays, despite the irrelevant color variation, was r=.69). For the color combinations the correlation in performance across display types was higher (r=.57). The quite low shape correlation suggests that different features of these form stimuli were probably involved in their discrimination when tested as geometric and texture displays. Several reasons for this divergence are possible. One is that densely packed texture displays more strongly activate global grouping process than do the loosely spaced geometric displays, a conclusion consistent with the human texture results described above. A second is tied directly to the different element sizes used in each display. Although it has been found that pigeons show a substantial degree of shape invariance across different sizes (Lombardi & Delius, 1990; Pisacreta, Potter, & Lefave, 1984), this is an obvious source for the difference that cannot be ruled out in this case. Clearly equating for element size in the two display types is needed to decide among these possibilities, we are also actively pursuing that goal. Nevertheless, despite this limitation, these current results represent the best available evidence that odd-item texture discriminations and odd-item visual search procedures tap different processes, as well as common ones, in pigeons

Next, we conducted hierarchical cluster analyses (average method) on each display type=s unfolded similarity matrix as derived from the accuracy results collected during the same 100 sessions of testing. Cluster analysis is one of several widely used multivariate techniques for detecting and modeling the psychological similarities present within a set of data. As such, the cluster diagrams in the figures below show the relative similarity of the shape elements when tested within each display type. These summary cluster analyses show the pattern for the twelve shapes and the twelve colors as tested with the geometric (top panel) and the texture (bottom panel) display types.

These analyses suggest that a major factor in both types of discriminations is whether the shapes are solid geometric forms or composed from multiple line segments. Testing pigeons with black figures on a white background, Blough (1990) has found the same fundamental division between filled and unfilled shapes (see also Blough & Blough, 1990). These congruent findings thus suggest that relative brightness, or a very closely correlated attribute, is an important component feature for pigeons in distinguishing forms, and quite likely surface properties, no matter how the displays are configured.

Closer inspection reveals other differences between the texture and geometric analyses that go beyond this factor. Only a handful can be highlighted. For instance, the Abow-tie@ shape strongly clusters with the other solid forms when presented as a texture stimulus, but shifts to the linear shape cluster when appearing as a geometric display. The AU@ shape also shifts, but in the reverse direction. Additionally, the angular components of the Achevron@ shape dominate its discrimination when tested as a textured stimulus, as reflected by its greater confusion with the pointed Astar@ shape, while in the geometric displays its horizontal aspect seems more important as evidenced by its greater confusion with the horizontal line. Collectively, these kinds of shifts show how the different configurations/sizes of these shapes change or emphasize different, and perhaps new emergent, features of these forms.

Cluster analyses were conducted also on the color data as well, and their results are shown in the two panels to the left. Reflective of the greater correlation in choice accuracy found among the two display types, both panels look similar. Their overall pattern shares much in common with the human perception of these colors, despite the considerable differences in the mechanisms of color vision in these two species. Strong clusters emerged for the Ared@ and Ablue@ ends of the spectrum, with the yellow/green values somewhere in between, a pattern consistent with published spectral sensitivity and color naming data for pigeons (Blough, 1957; Schneider, 1972; Wright & Cumming, 1971). There was a more variable clustering solution formed around the gray, coral, and pink values. While these color values differed considerably in their spectral composition, they were among the lighter of the stimuli that we tested, and their grouping may reflect the contribution of brightness to what is otherwise a hue discrimination. While the final solutions for each display type are highly similar, they are not identical. Just as texture density may alter form perception in pigeons, it may also alter the relative perception of the screen=s color values by means of brightness and color contrast (Hodos & Leibowitz, 1978; Varela, Palacios, & Goldsmith, 1993).