Fast exact local-mean based classification algorithm using class rejection and initial distance assigning processes

• Published in: Pattern recognition letters Vol. 33; no. 12; pp. 1507 - 1512
• Main Authors: Lin, Jun-Feng, Liaw, Yi-Ching, Chan, Kwan-Yee, Wu, Song-Chan
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.09.2012
• Subjects: Fast k-nearest-neighbor search; k-nearest-neighbor equality; Local-mean based classification; k-nearest-neighbor equality; Local-mean based classification; Fast k-nearest-neighbor search
• ISSN: 0167-8655; 1872-7344
• DOI: 10.1016/j.patrec.2012.04.013

► This paper proposed two processes to boost the local-mean based classification algorithm. ► The proposed method generates the same result as that generated by the local-mean based algorithm. ► The proposed method can effectively reduce the computation time of the local-mean based algorithm. The local-mean based classification (LMC) algorithm is an effective classification method. It can reduce the influence of outliers and can achieve better result than most classification algorithms. To classify a test sample with unknown class using the LMC algorithm is very time consuming. To overcome this problem, this paper presents a fast exact LMC method to reduce the computation time of the LMC algorithm using a class rejection process. When the proposed method cooperates with fast nearest k neighbors finding algorithms, an initial distance assigning process is also applied to improve the performance of rejecting impossible samples. Experimental results show that the proposed method can effectively reduce the computation time of the LMC method and can improve the capability of rejecting impossible samples when a fast kNN search method is applied to the LMC algorithm.