Optimising Statistical Shape Models Using a Minimum Description Length Approach

Statistical shape models are used widely as a basis for segmenting and interpreting images. A major drawback of this approach is the need to identifying corresponding points across a training set of segmented shapes. By posing the problem as one of minimising the description length of the model, we describe an efficient method that automatically defines a correspondence across the set of shapes. As the correspondence does not use an explicit ordering constraint, it generalises to 3D shapes. Results are given for training sets of 2D boundaries, showing the automatic method constructs better models than ones built by hand.