Combining modified Nearest Neighborhood and fuzzy K-Nearest Neighborhood methods for classification of urban areas (Case study: Shahriar, Tehran, Iran)

1.646 405


Abstract. In this paper a method for combining classification of NN[1] and fuzzy k-NN[2]is represented. At first for reducing computation costs and noise elimination, some parts of training data which have the least value of dependency function, have been omitted. For classifying with modified NN method, an area is defined for each class, so that it comprises the entire training data belong to that class and the minimum possible amount of other classes. Pixels which have been in a defined area of one class would be labeled in that class and also pixels which have been in a defined area of more than one class would be labeled with fuzzy k-NN method. The main purpose of this task is to reduce computation costs compared to k-NN method and also to increase precision compared to NN method. This method (combining modified NN and fuzzy k-NN) is evaluated by data with 5 classes and it is been observed that computation time and classification precision have been improved by 14% and 2% respectively. 


remote sensing, combining classification methods, modified Nearest Neighborhood, k- Nearest Neighborhood, fuzzy logic

Full Text:



Cover, T.M., Hart, P.E., Nearest neighbor pattern classification, Transactions on information theory, (1967), pp 21-27

Fix, E., Hodges, J.L., Discriminatory analysis nonparametric discrimination: small sample performance, project 21-49-004, (1952)

Fukunaga, K., Mantock, J.M., Nonparametric data reduction, transactions on pattern analysis and machine intelligence, (1984), pp 115-118

Joswik, A., A learning scheme for a fuzzy k-NN rule, pattern recognition letters, (1983), pp 287-289

Joswik, A., Roli, F., Dambra, C., A multistage synthesis of modified NN rule and its application for remote sensing images, prot IPTA (1993) pp 435-438

Serpico, S., Joswik, A., Roli, F., A parallel network of modified 1-NN and k-NN classifiers: application to remote-sensing image classification, Pattern recognition letters, (1998), pp 57-62

Robert A. Schowengerdt, 2007, Remote sensing Models and methods for image processing, Third Edition, pp558