We plan on building an algorithm to automatically convert data visualizations, such as graphs and charts, into raw data. Sometimes, useful information is only available graphically, but charts are much more difficult to manipulate than the data the chart is based on. Additionally, visualizations are often poorly designed. If we could convert images into data, we could... something cool.

Our current plan is to build the core of this algorithm on the experts framework. For example, one of the first tasks when presented with an arbitrary data visualization is to determine which type of visualization it is: pie chart, scatterplot, time series, and so on. By creating an "expert" for each type of visualization, we can analyze the image to determine which expert best matches the image provided. This approach can be extended into the subtasks of the project, such as matching labels to relevant graph areas. We plan on starting by attempting to identify a small number basic graph types produced by Excel, since those are both common and predictable in design. Depending on the success of our algorithm, we can extend the scope of what can be analyzed to include more varieties of visualizations.