Analysis Chains

As mentioned earlier, analysis modules can be strung together into analysis chains. An analysis chain is simply a directed acyclic graph (DAG), where the nodes of the graph are analysis modules. The edges of the graph are data-dependency links, which specify that the output of one module in the chain should be presented as inputs to another module.

The main user operation provided by the analysis engine is to execute an analysis chain against a dataset of images. A chain must be well-formed before it can be executed. A chain is well-formed if all of the following criteria are met:

1. Each link in the chain connects nodes that belong to that chain.
2. The types of the output and input of a link must match — they must either be identical semantic types, or the output must be untyped.
3. There cannot be any cycles in the analysis chain.
4. There cannot be more than one link connected to any single formal input in the chain.

Every formal input in the chain is either free or bound. The bound inputs are those which have links connected to them. The free inputs do not. Since every formal input in a module is guaranteed to have a value, each free input must have its value provided by the user in order for a chain to be executed. Note that this value can be an empty list of attributes, but this must be specified explicitly.