Ari Nieminen

Algorithms for real-time trend detection

Abstract In this paper, we develop FIR median hybrid (FMH) filter based algorithms for real-time trend detection. The algorithms are designed so that trends are gradually refined as data becomes available. Special attention is paid to the clear detection of sharp changes in trends. The algorithms are based on a new concept of five point median operations taken over the outputs of linear subfilters or some other auxiliary outputs. The noise attenuation and the edge preserving abilities of several FMH trend detection filters are analyzed. The results show clearly the usefulness of the FMH trend detection. Moreover, the in-place growing FMH filter based trend detector was found to have significant advantages over the FMH based methods. The trend filtering concept can also be successfully applied to the filtering of the beginning and end of a finite length data sequence.

Comments on "Theoretical analysis of the max/median filter" by G.R. Arce and M. P. McLaughlin

In the original paper (ibid., vol.ASSP-35, no.1, p.60-9, Jan. 1987), an approach (the max/median filter) to remove noise from images without distorting geometrical information was introduced. The proposed filter structure, due to its asymmetrical nature, is able to preserve only the subtle signal structures that are larger in intensity than the local background level. The commenters suggest another filter structure that has the ability to preserve both positive and negative geometrical information with respect to the background level. >

Motion field restoration using vector median filtering on high-definition television sequences

A nonlinear algorithm called vector median is introduced for motion-compensated upconversion of high-definition television (HDTV) images. The postprocessing technique of vector-median filtering is defined theoretically and applied to the HD-MAC HDTV coding system to compare sample motion-vector fields before and after the process. The quantitative effect of the vector-median operation with different test sequences is given as the rate of motion vectors changed by postprocessing, and upconverted images are shown to improve significantly with postprocessing