The SmartROI Tool is designed to support the automatic location of arbitrarily complex polygonal boundaries in grey level image data to an average spatial accuracy of about one pixel. The software supports a range of techniques covering conventional snakes and the more sophisticated deformable template models. The specific choice of algorithm applied can vary between these extremes allowing either shape constraints with a conventional location potential for simple tasks, or averaged or covariant models of grey level profiles for more complex tasks. These techniques are based on eigen-vector approximation to sample data sets. The most appropriate (and most accurate) technique will be problem specific and a separate model will need to be built and tested for each task. The model construction process is often manually intensive, but once a trained model has been developed they can be applied automatically and are often accurate and reproducible as a way of identifying regions in images. These techniques have thus found considerable use in recent image processing tasks.
In order to provide a simple but generic approach it is necessary to make some assumptions regarding the expected model and its construction. The basic assumption here is that the model to be located is a simple region consisting of an inner and/or outer portion and that there is a preferential orientation axis and overall scale. We also assume that sensible landmarks can be located, either along perpendicular lines at regular intervals along the axis or radially. This model has been chosen on the basis of simplicity and has a surprising range of applicability for the majority of basic image location tasks. In particular, the use of a main axis and perpendicular line mark-up considerably simplifies the representation of the shape model by ensuring that unwanted errors in the 2D location of the set of points are efficiently removed in the eigen vector model.
The number of allowable degrees of freedom is limited to ten for simplicity of interface design, though the number required for most tasks is often considerably smaller than this. Once an object has been located the region of interest defined by the polygonal boundary is then available for use in any Tv, including the imcalc Tv, thus allowing simple region extraction via the Imcalc Tool roi facility for further processing and automated data extraction.