Characteristic Views

The concept of a characteristic view (CV) is useful in appearance-based object recognition [5], although here we choose to treat the concept in a very loose, naturalistic fashion, rather than the principled but restrictive definition of Freeman and Wang.

Characteristic views are intended to help us obtain a representative and adequate grouping of views, such that a given level of recognition accuracy may be achieved using the minimum number of stored views [16]. Clearly, this has important implications for the storage space needed to represent each object, and the number of matches which must be performed at run-time for the purpose of recognition. View grouping has been addressed using CVs and aspect graphs (AG). An aspect graph [7] enumerates all possible appearances of an object, and the change in appearance at the boundary between different aspects is called a visual event.

However, aspect graphs grow to unwieldy sizes for complex, non-polyhedral objects, since all visual events are considered sufficiently important to define a new boundary between aspects[12]. It is difficult to define a single face when an object is composed of piecewise curved surfaces[11]. Even slight changes in viewpoint may result in more of the curved surface(s) either coming into, or disappearing from, the view. Thus, either the size of the aspect graph must be controlled using appropriate heuristics [15], or a less rigid approach considered. We choose to adopt the latter course, and treat the concept of a characteristic view in a more psychophysical manner, as a natural groupings of views.

A possible method of identifying natural CVs, in this sense, is to use clustering to identify natural view groupings [9]. From a human perspective, all views of an object which form a CV should ``look'' more similar to each other than to any view from a different CV. If all the views within a CV are similar, then only one such view (or an average view) need be stored and matched for recognition. It follows that the larger, on average, each CV is, the fewer model views need be stored in order to span the view-sphere, and the more efficient both the learning and recognition of objects will become.

The representation used for the model views has great influence upon the average extent of the CVs. A representation which is relatively stable over a range of viewpoints will result in larger CVs, on average, than one which changes greatly for small shifts in viewpoint. However, this local invariance must not be at the expense of loss of detail, since this will impair the ability to discriminate between objects. Note that the concept of a characteristic view extends to variations in light source, so that we can also group a set of light source directions together so that a given group of views and light source directions result in similar images.