[top of page 13 of handout]
Van Hamme, L.J., Wasserman, E.A. and Biederman, I. (1992)

  • This was a modification of an experimental paradigm originally used to assess the role of very local features in human object recognition.

  • Pigeons were trained with 12 copies of each of 4 modified line drawings: an elephant, a mushroom, a chair and a telephone.

  • The modification was that half of the original outline was deleted.

  • 4 birds were used, each being shown 48 stimuli each session.

  • The birds had to peck a small viewing screen (7x7 cm) when the drawings were projected on it.

  • After 30 pecks, choice keys outside each of the 4 corners of the viewing screen were illuminated.

  • Only one of the 4 choice keys was correct for each of the stimulus categories. If the correct choice was made the pigeon received 2.5 sec access to grain.

  • After an incorrect choice no grain was presented.

  • After training (60 days) there was a test phase in which the modified drawings used were sometimes replaced by the other half of the full line drawing

  • (see here for examples).

  • The result was that all 4 birds were significantly better than chance at making the correct response to these new versions of the drawings, the precise details of which they had never seen before.

  • The mean percent correct during the test phase was 77% correct for the drawings used in training and 67% correct for the new drawings. (Chance is 25% correct).

  • There was thus substantial generalization to the new drawings.

  • In further experiments, scrambled versions of the stimuli, were less well discriminated.

     

    [middle of page 13 of handout]
  • The authors conclude that a particular theory of object recognition in humans – “Recognition by components” applies to some extent also to pigeons.

 

 

 

 

 

 

 

 

 

[ see intranet and page 14 of handout]
Effects of geon deletion, scrambling, and movement on picture recognition in pigeons
Kirkpatricksteger_K, Wasserman_EA, Biederman_I
Brown Univ,Dept Psychol,Box 1853,Providence,Ri,02912 Univ Iowa,Dept Psychol,Iowa City,Ia,52242
Journal of Experimental Psychology-Animal Behavior Processes, 1998, Vol.24, No.1, Pp.34-46
  • This paper describes an experiment using a very similar procedure to that described above.

  • Discrimination training was was with 4 objects, each of which contained 4 geons. all were same height 2.9 cm each drawing appeared in 4 different quadrants of the screen (necessary from pilot work)

  • Each session was 10, 16 trial blocks, each 16 was 4x4 (at random). Requirement was 30 pecks at the display. If choice correct (of 4) then one or 2 45 mg food pellets were given as rewards.

  • Only the first choice was counted for data, but there was a correction procedure: a trial was repeated until the correct response was made.

  • The result of transfer tests was that scrambling or moving the individual components of pictures reduced the accuracy of performance. This is not terribly surprising, but it demonstrates that the birds were learning something more in their initial training that just to recognize isolated features of the displays.

  • Exampes of the changes made to the stimuli are given here.  

     

     

    Abstract. E. A. Wasserman, K. Kirkpatrick-Steger, L. J. Van Hamme, and I. Biederman (1993) demonstrated that scrambling an object's parts or "geons" (I. Biederman, 1987) produced marked reductions in the pigeon's picture recognition accuracy, indicating that discriminative responding to pictures is controlled by more than simple particulate features. The present effort was designed to further assess the contribution of various stimulus attributes to picture perception. Four pigeons were trained to discriminate 4 line drawings of human–made objects. Subsequent tests revealed that (a) the spatial organization of the geons was a major contributor to picture recognition; (b) the individual geons were also important, with different pigeons demonstrating control by different subsets of geons; (c) recognition of the training drawings was positionally invariant; and (d) the points where the geons contacted one another were largely unimportant for picture recognition. The results provide further support for the notion that pigeons perceive both global and local aspects of complex stimuli in much the same way as do humans.

     

     

     

     

     

     

    Conclusion

    The results provide further support for the notion that pigeons perceive both global and local aspects of complex stimuli in much the same way as do humans.