Picture-object recognition in a comparative approach: performance of humans (Homo sapiens) and pigeons (Columba livia) in a rotational invariance and a complementary information task

摘要

Pigeons and humans are two highly visual species that have evolved separately for udabout 310 million years (Kumar and Hedges, 1998) and developed largely convergent visual udsystems due to similar visual needs. To investigatepigeon vision and cognitive abilities twodimensional pictorial stimuli are often used. However, it is not entirely clear, how pigeons udperceive such stimuli and whether or not they can associate photographs with real objects. udIn the present study nine pigeons and eleven humans were trained to discriminate udbetween photographs of two biologically irrelevant objects (“Greebles”). The pigeons were udhoused in an aviary containing the real Greebles and were trained in wooden chambers where udthey had to peck on a Plexiglas disk when positive stimuli were presented, thus obtaining udfood. Humans were trained with the same stimuli presented on a computer screen and had to udclick with a computer mouse on positive stimuli. Results showed that humans were much udfaster at learning to discriminate the two Greebles. In the first test, pigeons and humans had to uddiscriminate new rotational views of the Greebles. Humans performed equally well on udinterpolated test views (i.e. views that lay between the training views) and extrapolated views ud(i.e. views outside of training range), while pigeons performed better on interpolated than on udextrapolated test views. Therefore, it can be concluded that object recognition was viewpointindependent for humans and viewpoint-dependent for pigeons. In the second test, following a udprocedure by Aust and Huber (2006), pigeons were presented with parts of the Greebles that udwere not included in training and the first test to see whether they formed associations udbetween the 2D images and the 3D objects in their aviary. They did not discriminate these udparts correctly. The test was repeated with three of the test views already used in the second udtest but presented in different sizes. Discrimination seemed to depend on the visibility of the udappendages and might have been based on visual features of the pictures themselves without ud71 udrecognition of what they portrayed. The results of this study were compared to a previous udstudy in which pigeons were trained to discriminate either real Greebles, holograms, or udcomputer images of them. There, too, the real Greebles were installed in the pigeons’ aviary; udhowever, the pigeons trained and tested on computerimages lived in the adjacent aviary and udthus only had limited visual contact to them. We wanted to find out whether the more udextensive visual contact to the Greebles had any influence on the pigeons’ performance. udHowever, there was no difference in performance between the two groups. This is evidence udthat the result of the previous study — better performance with real objects and holograms udthan with computer images — was not based on the fact that pigeons trained with the latter udstimulus type had only limited visual access to thereal 3D objects.