Knowledge based adaptive fuzzy strategy for target of interest differentiation

This chapter mainly discusses the issue of target of interest recognition based on image. To implement the function, we present a fuzzy knowledge based system for target of interest differentiation with its application in matting enhancement as a showcase. How to use human experience as knowledge to develop an adaptive strategy has been addressed for optimization. A recursive optimization framework is used in the scheme with adaptive fuzzy learning strategy. Considering the uncertainty of data and processing, the proposed scheme successfully applies the expert human knowledge into matting. The uncertainty of data and solution are both taken into account. The measurement involved with the energy and the dimension of the anonymous pixels is introduced into the objective for optimization. The key to the success of enhancement of matting lies in the recursive learning with kernel learning and adaptive fuzzy updating strategy, enabling the solution preserve more desired details while filtering the untrustworthy information. With the consideration of the uncertainties involved with color condition, priori knowledge and solution, tail recursion and uniqueness of optimization guarantee the stability of the matting solution, enabling the proposed scheme robust to various uncertainties in matting. Experiment results are given to demonstrate the effect of the proposed method with the comparison between before and after enhanced matting. The analysis and evaluation are given as well.