Description Evidence

This relationship gives direct evidence for an object, because all properties of the description type (i.e. a generic objects) are also object properties - the type is a description of the object. The object may have several description types: a "square" is both a "rectangle" and a "rhombus". Hence, evidence for a description type is also evidence for an object that uses the description, and this is expressed by a plausibility relationship.

Constraints that help specify the description evidence computation are:

• The more plausible the description, the more plausible the object.
• Each description gives additional evidence for an object.
• A weight expresses the importance of the description relative to other property evidence.
• The context of the description is that of the object.

Formally, the description evidence computation is defined:

Given:
	a model instance of type $M$ in image context $C$
	a set {($D_i,w_i$)} of descriptions $D_i$ of $M$ with associated weights $w_i$
	a set {$p_i$} of plausibilities of description $D_i$ in context $C$
Then, using the modified harmonic mean, the description evidence is:
	$evd_{desc} = harmmean(\{(p_i,w_i)\})$
The resulting weight of evidence is:
	$w_{desc} = \sum w_i$

If no description types are defined, this computation is not applied. The portion of the network associated with this evidence is similar to those shown above.