Chapter 5: Realizing a Metadata Framework 169 The geometrical fragment specifi cation is in the form #img:rect(left,top,right,bottom[,depth]) where left, top, right, and bottom defi ne rectangular fragment of the object. Depth can be used for three-dimensional objects. Similarly to time-based fragments, left and top are inclusive, whereas right and bottom are not. This also means that right and bottom can extend over the dimensions of the geometrical object. Other geometrical shapes can be also used, but only img:rect is currently defi ned, for example, http://some.server.example/images/ex.jpg#img: rect(0,0,200,300) will defi ne a top left fragment of image ex.jpg that is 200 pixels wide and 300 pixels tall. 5.6.10.4 Embedded Objects Objects can be embedded in two ways, by linking or by real embed- ding. If the objects are embedded using links, the parent object does not contain the other object, but just a link to it. In this case, we do not need fragment URIs to embed the object, since it must have an ID of its own and a relationship can be used instead. Where the embedding is real, such as in the case of an audio track in a video clip, we need fragment URIs. The fragment URI specifi cation for embedded objects is based on an object array. Embedded objects are indexed from the beginning of the document. The fi rst index is 0. The fragment specifi cation for embedded objects is therefore in the form #object[index] where index is an index to the embedded object, for example, fi le://docs/word.doc#object[5] refers to the sixth embedded object in word.doc. 5.6.11 Extending the Ontology As discussed earlier, it is almost impossible to model everything that any software developer may invent in the future. For that reason the