Feature Fitting

When data is extracted from in image it is often converted into a more convenient abstract, higher level, representation. This is generally done by fitting a model to the data. Several problems exist, in particular robustness and efficiency. Under noisy conditions many fitting methods fail, while complicated non-linear models may be difficult to fit and require slow iterative techniques. Work is underway on developing new methods for fitting ellipses and superellipses to data. For the former this involves applying robust statistics to improve robustness. For the latter new error functionals have been developed as well as new simpler, faster fitting methods. An example of fitting a superellipse to the data (in red) is shown below. With so much data missing it is extremely difficult to get a reasonable fit.

superellipse

More details are given in:

You can download code to implement several ellipse fitting methods as well as code to implement several superellipse fitting methods.

return to Paul Rosin's homepage