2008.Adaptive local linear regression with application to printer color management电子书.pdf

936 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 17, NO. 6, JUNE 2008 Adaptive Local Linear Regression With Application to Printer Color Management Maya R. Gupta, Member, IEEE, Eric K. Garcia, and Erika Chin Abstract—Local learning methods, such as local linear regres- regression estimate is bounded by the variance of the measure- sion and nearest neighbor classifiers, base estimates on nearby ment noise. training samples, neighbors. Usually, the number of neighbors We apply our proposed adaptive local linear regression to used in estimation is fixed to be a global “optimal” value, chosen by cross validation. This paper proposes adapting the number of printer color management. Color management refers to the task neighbors used for estimation to the local geometry of the data, of controlling color reproduction across devices. Many com- without need for cross validation. The term enclosing neighbor- mercial industries require accurate color, for example, for the hood is introduced to describe a set of neighbors whose convex production of catalogs and the reproduction of artwork. In addi- hull contains the test point when possible. It is proven that en- tion, the rising ubiquity of cheap color printers and the growing closing neighborhoods yield bounded estimation variance under some assumptions. Three such enclosing neighborhood definitions sources of digital images has recently led to increased consumer are presented: natural neighbors, natural neighbors inclusive, demand for accurate color reproduction. and enclosing k-NN. The effectiveness of these neighborhood Given a CIELAB color that one would like to reproduce, t