Avian Visual Cognition

Home Page

Next Section:
Importance of Components


Spatial Organization of Components

A number of studies have recently demonstrated the importance of organizational properties of object components to picture recognition in avian species (e.g., Brown & Dooling, 1993; Kirkpatrick-Steger, Wasserman, & Biederman, 1996, 1998; Wasserman et al., 1993; Watanabe & Ito, 1991). The primary approach in these studies has been to train birds to recognize particular intact objects and then test with scrambled versions of those objects. For example, pigeons might be trained to discriminate among line drawings of four objects such as a watering can, an iron, a desk lamp, and a sailboat. Following, training, the pigeons would then receive tests in which the object components would be spatially rearranged, such as in the example below.

The scrambled versions to the right resemble the intact objects because they contain the same parts. However, to the human eye, they are also discriminably different from the original objectOne scrambled version of each of the four training objects. All four geons were spatially reorganized so that the original interrelations were no longer present. because the organization of the components is different. The following set of studies examined Click here to see the scrambled versions of the stimuli whether pigeons, too, would be able to discriminate the scrambled verions of the objects from the intact verions. The pigeons were tested with several different scrambled versions of each object. 

According to Particulate feature theory (Cerella, 1986), pigeons should not be able to recognize a scrambled version as different from an intact version of an object because the scrambled and intact versions contain the same features (see Charlie Brown example). On the other hand, Recognition-by-components (Biederman, 1987) postulates that scrambled versions should be discriminated from the original objects because the geon interrelations are altered by the operation of scrambling.

Three different experiments using different pigeons (Wasserman et al., 1993; Kirkpatrick-Steger, Wasserman, Biederman, 1996, 1998) involved tests with scrambled versions of the watering can, the iron, the desk lamp, and the sailboat. In all three studies, theClick Here to see results of the tests with scrambled figures original drawings were discriminated above 80% correct at the end of training. Performance to the scrambled versions was significantly poorer than the performance to the original versions in all three studies. However, performance to the scrambled versions consistently exceeded the chance level of 25%. The level of performance to the scrambled versions is consistent with the notion that the pigeons were sensitive to the change in the spatial organization of the components, but they must have still recognized some features of the original objects because of the above-chance performance to the scrambled versions.

One interpretation of the detrimental effect of scrambling on recognition accuracy is that the scrambled versions contained different points of contact between the geons. An adept particulate perceiver might be able to discriminate the change in contact points between the intact and scrambled objects. Kirkpatrick-Steger, Wasserman, and Biederman (1998) conducted tests to assess the contribution of the geon intersections to object recognition. Following training on the four-key choice procedure to discriminate among the intact versions (Connected-Original, CO) of the watering can, the iron, the desk lamp, and the sailboat, pigeons were tested with  three different types of stimuli in addition to the original training objects. One type of test drawing, the Disconnected-Original (DO Click Here to see the three different types of training stimuli) drawings, contained the four geons in their original spatial organization, but the geons were moved apart slightly, so as to remove their intersections. A second type of test stimulus, the Connected-Scrambled (CS) drawings, were scrambled images shown above with novel spatial organizations and altered geon intersections. The third type of test stimulus, the Disconnected-Scrambled (DS) drawings, were the same scrambled images, but with the geons moved apart slightly to remove any information about the change in geon intersections due to the operation of scrambling. If the alterations in the geon intersections were responsible for the scrambling effect, then one would expect similar levels of performance on the three test stimuli.

Disconnection of the geons in the original spatial organization (DO) resulted in accuracy scores that did not differ from the original objects (CO). The connected-scrambled (CS) andClick Here to see the results of the geon interaction test disconnected-scrambled (DS) drawings were discriminated at a similar level of accuracy, which was significantly poorer than the original drawings. The test results indicate that the geon intersections were not a major contributor to object recognition. These results are consistent with a study by Van Hamme, Wasserman, and Biederman (1992). They trained pigeons to discriminate four objects in which half of the contours and vertices were deleted. Then pigeons were tested with complimentary contours which were in an intact or scrambled arrangement. The complementary images were discriminated at a high level of accuracy, but the scrambled contours were recognized much more poorly. Their results indicated that the contours and geon intersections alone were insufficient to account for successful object recognition.

Because scrambling had a significant effect on recognition accuracy that could not be accounted for by the changes in the geon intersections, it appears that pigeons are sensitive to the spatial organization of object components. This sensitivity should then allow pigeons to discriminate between objects that are composed of the same components, but with different organizations. (see the cup vs. pail example).

In order to test this proposition, pigeons were trainClick here to see scrambled versions of the stimulied to discriminate four different scrambled versions of an object (Wasserman et al., 1993). Bird 1 was trained with the four-key choice procedure to discriminate among scrambled versions of the watering can; bird 2 was trained with scrambled versions of the iron; bird 3 was trained with scrambled versions of the desk lamp; and, bird 4 wasClick here to see the results from the learning test trained with scrambled versions of the sailboat. All four pigeons learned to discriminate among the four scrambled versions of an object. The degree of accuracy at the end of training (mean of four pigeons = 75.9% correct) was similar to accuracy levels obtained with pigeons that were trained to discriminate among the intact training objects (mean of four pigeons = 80.5% correct). The results provide further evidence that pigeons are sensitive to the spatial organization of object components.

All of the above results indicate that pigeons are sensitive to the spatial organization of object components. This sensitivity is inconsistent with Particulate feature theory (Cerella, 1986). Because Recognition-by-components (Biederman, 1987) proposes that both the local features (geons) and their spatial organization are important for recognition, an additional demonstration that geons also contribute to object recognition in pigeons would provide substantial support for the application of RBC to object recognition in pigeons as well as humans. Evidence from studies where geons were deleted or moved (Kirkpatrick-Steger, Wasserman, and Beiderman (1998)) indicate that pigeons may be sensitive to an object's geons as well as their spatial organization.

Next Section: Importance of Components: Geon Deletion and Movement