A Fast and Efficient K-Nearest Neighbor Classifier Using a Convex Envelope

• Published in: Recent Trends in Image Processing and Pattern Recognition Vol. 1576; pp. 320 - 329
• Main Authors: Yepdjio, Hermann, Vajda, Szilárd
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2022; Springer International Publishing
• Series: Communications in Computer and Information Science
• Subjects: Character recognition; Classification; Convex envelope; Digit recognition; K-nearest neighbor
• ISBN: 3031070046; 9783031070044
• ISSN: 1865-0929; 1865-0937
• DOI: 10.1007/978-3-031-07005-1_27

In this paper, we propose a fast and efficient method to classify all kinds of patterns using the classical k-nearest neighbor (kNN) classifier. The kNN is one of the most popular supervised classification strategies. However, –for large data collections, the process can be very time consuming due to the tedious distance calculations. Our aim is to provide a generic strategy for all kinds of data collections by calculating fewer distances as in the classical approach. For that reason we propose a data selection technique that reduces the original data to a limited one which contains only some class prototypes. The prototypes are representatives of each class and are selected based on the notion of convex envelope. The experiments on multiple benchmark data collections such as MNIST, Fashion-MNIST and Lampung characters show a considerable speed up (up to 12x) in the classification, while reporting similar or slightly less classification figures than the classification results obtained using the complete data.