Recursive square-root filters

Abstract

This paper presents a derivation of recursive square-root filters, which can update either a symmetric positive data gathering matrix or its inversion. These filters differ from usual recursive approaches by their ability not only to supply new data to these matrices but also to simultaneously remove data previously fed to these matrices during their build-up. A condition for stable data removal is proven. Efficient recursive statistics of the least square or the maximum pseudo-likelihood type can be built using these results as it is briefly demonstrated on the colour texture segmentation example. © 2000 IEEE.