Viewpoint Invariance in Object Recognition

This article reviews a number of different approaches to the study of the way the visual system organises and recognises objects and shapes in the natural environment. A brief critical review of these different approaches and their impact on current theories of objects recognition are discussed. Recent approaches have argued that objects are represented in memory by a number of characteristic views and that recognition proceeds by matching the input to the nearest (or best fitting) stored view. Disagreement, however, lies in which process the visual system uses to match the input with the representation. Some have argued that the input is mentally rotated to match the nearest stored view and others argue that the visual system interpolates between the stored views. Recent experimental work favours the mental transformation approach. The results are not conclusive, however, and it could be argued that a non-linear interpolation approach may prove to be the best model of the behaviour of the visual system when recognising shape. These different approaches are compared and recent work in our laboratory is discussed in light of the current models. Our findings support the idea that the visual system stores characteristic views of objects and that unusual views are matched to the nearest stored view. More research is necessary, however, in order to determine the nature of this matching process.