Controlled condensation in K-NN and its application for real time color identification

 

Guardado en:
Detalles Bibliográficos
Autores: Villar-Patiño, Carmen, Cuevas-Covarrubias, Carlos
Formato: artículo original
Estado:Versión publicada
Fecha de Publicación:2017
Descripción:k-NN algorithms are frequently used in statistical classification. They are accurate and distribution free. Despite these advantages, k-NN algorithms imply a high computational cost. To find efficient ways to implement them is an important challenge in pattern recognition. In this article, an improved version of the k-NN Controlled Condensation algorithm is introduced. Its potential for instantaneous color identification in real time is also analyzed. This algorithm is based on the representation of data in terms of a reduced set of informative prototypes. It includes two parameters to control the balance between speed and precision. This gives us the opportunity to achieve a convenient percentage of condensation without incurring in an important loss of accuracy. We test our proposal in an instantaneous color identification exercise in video images. We achieve the real time identification by using k-NN Controlled Condensation executed through multi-threading programming methods. The results are encouraging.
País:Portal de Revistas UCR
Institución:Universidad de Costa Rica
Repositorio:Portal de Revistas UCR
Lenguaje:Español
OAI Identifier:oai:portal.ucr.ac.cr:article/22354
Acceso en línea:https://revistas.ucr.ac.cr/index.php/matematica/article/view/22354
Access Level:acceso abierto
Palabra clave:supervised classification
nearest neighbours
multi-threading
condensation
prototype selection
clasificación supervisada
vecinos cercanos
programación multihilos
condensación
selección de prototipos