Design of robust types of mode filter for signal and image processing applications

This paper considers filters which can be used for mode estimation. It starts by studying a method which has been used for mode filtering of images, and then goes on to a more systematic analysis covering the rapidly advancing field of robust statistics. It demonstrates the importance of clarifying the specification of the required filter. It has listed five parameters that are important in the design of peak location filters, and has also shown some of the tradeoffs that exist between them. The need to adapt a filter in an appropriate way to cope with specific signals has been demonstrated: there remains the possibility of making filters self-adapting in this way. Finally, the paper provides cogent arguments against the LMedS approach being the `ultimate´ method of mode estimation