Culik and Karhumaki studied weighted finite automata (WFA) as devices
computing real functions. The main motivation was to give specificatio
ns of graytone images as local grayness functions. Culik and Kari gave
an algorithm for automatic image encoding using WFA as a basis for a
practical image data compression method. In this paper we introduce k-
tape weighted finite automata. We are mainly interested in the case of
2 tapes called weighted finite transducers (WFT). We show that the mo
st commonly used image transformations can be defined by WFTs. We also
show that WFT transformations are closed under union and composition
and that the family of WFA images is closed under WFT transformations.