Improved LBF model combined with fisher criterion

Abstract

An improved local binary fitting (LBF) [1] model combined with Fisher criterion is introduced in this paper. We hope to improve collection degree within classification and separation degree between two classifications, pixels are distributed over local domain of two sides of the boundary. Based on this idea, a new energy term are constructed. The new model is more robust under global information of pixels in the local region of boundary. Furthermore, the narrow band implementation of new model is able to reduce the computational cost greatly. © 2012 Springer-Verlag GmbH.