This paper attempts to view texture analysis and synthesis in image processing as a problem of realization and subspace identification in terms of reciprocal processes defined on a finite interval. We discuss modeling of textures in the framework of acausal stochastic systems and stochastic realization theory.