Automatic localisation of correspondences for the construction of Statistical Shape Models from examples has been the focus of intense research during the last decade. Several algorithms are available and benchmarking is needed to rank the different algorithms. Prior work has focused on evaluating the quality of the models produced by the algorithms by measuring compactness, generality and specificity. In this paper severe problems with these standard measures are discussed. We propose that a ground truth correspondence measure (gcm) is used for benchmarking and in this paper benchmarking is performed on several state of the art algorithms. Minimum Description Length (MDL) with a curvature cost comes out as the winner of the automatic methods. Hand marked models turn out to be best but a semiautomatic method is shown to lie in between the best automatic method and the hand built models in performance.

Most implementations of single character recognition use standard arclength parameterization of the handwritten samples. A problem with the arclength approach is that points on curves of different samples from one character class may then actually correspond to parts of different structural significance. Many methods such as DTW and HMM have been successful partly because they are less sensitive to parameterizational differences. Given a sufficiently fine decomposition of a character sample into smaller segments, the complex non-linear variations of handwritten data can be viewed as a set of local linear transformations of the segments. In this paper we present a parameterization technique that implicitly defines such a structural decomposition. Experiments reveal that recognition rates for kNN template matching increase for reparameterized samples thus proving that the new parameterization removes redundance in a way that is genuinely beneficial for discrimination purposes. In addition to these quantitative results, visual inspection of the modes of singular value decomposition of reparameterized samples show that the new parameterization reduces the impact of parameterizational differences in shape variations of character samples.

Recently there has been much attention to MDL and its effectiveness in automatic shape modelling. One problem of this technique has been the slow convergence of the optimization step. In this paper the Jacobian of the objective function is derived. Being able to calculate the Jacobian, a variety of optimisation techniques can be considered. In this paper we apply steepest descent and show that it is more efcient than the previously proposed Nelder-Mead Simplex optimisation.

Automatic construction of shape models from examples has been the focus of intense research during the last couple of years. These methods have proved to be useful for shape segmentation, tracking and shape understanding. In this paper novel theory to automate shape modelling is described. The theory is intrinsically defined for curves although curves are infinite dimensional objects. The theory is independent of parameterisation and affine transformations. We suggest a method for implementing the ideas and compare it to minimising the description length of the model (MDL). It turns out that the accuracy of the two methods is comparable. Both the MDL and our approach can get stuck at local minima. Our algorithm is less computational expensive and relatively good solutions are obtained after a few iterations. The MDL is, however, better suited at fine-tuning the parameters given good initial estimates to the problem. It is shown that a combination of the two methods outperforms either on its own.