H. Crysandt (Germany)
Nearest Neighbor Search, Gaussian noise, Euclidian distance, Mahalanobis distance, Choleski factorization
Nearest Neighbor Search with Euclidian distance is a well known classification technique. (Most people know the Euclidian distance from school as Pythagoras' theorem). Thereby the distances between a feature vector of a query and the feature vectors of all classes in a database are measured and compared using the Euclidian distance. This paper describes how to increase the performance of the nearest neighbor classification algorithm when the distortion of the feature vectors of each class is known or can be estimated. If the distortion can approximately be described as additive correlated Gaussian noise other dis tances can be defined which perform better.
Important Links:
Go Back
