Ensemble-Based Segmentation in the Perception of Multiple Feature Conjunctions

Although objects around us vary in a number of continuous dimensions (color, size, orientation, etc.), we tend to perceive the objects using more discrete, categorical descriptions. For example, in the variety of colors and shapes on a bush, we can see a set of berries and a set of leaves. Previously, we described how the visual system transforms the continuous statistics of simple features into categorical classes using the shape of distribution. In brief, “sharp” distributions with extreme values and a big gap between them are perceived as “segmentable” and as consisting of categorically different objects, while “smooth” distributions with both extreme and moderate features are perceived as “non-segmentable” and consisting of categorically identical objects. Here, we tested this mechanism of segmentation for more complex conjunctions of features. Using a texture discrimination task with texture difference defined as length-orientation correlation, we manipulated the segmentability of length and orientation. We found that observers are better at discriminating between the textures when both dimensions are segmentable. We assume that the segmentability of both dimensions leads to rapid (within 100-200 ms, as our data show) segmentation of conjunction classes which facilitates the comparison between the textures containing these classes.