An increasingly large amount of texture features has been developed and used in combination with spectral features for classification problem. The question is how to select the combination/subset of texture features with spectral features best suited to a particular classification problem. In this paper, a modification of NarendraFukunaga algorithm ( 1977) is developed to select a global best subset of features. The computational aspect of the algorithm is discussed. The modified algorithm ensures to select a global best subset of features which was not identified by Narendra-Fukunaga algorithm. The experiment results also demonstrate that the modified algorithm is approximately 1 to 9 times more efficient in computational aspect than the Exhaustive Search method.

No abstract

No abstract

No abstract