KD-Tree and k-Nearest Neighbor Image Classification

Faculty Mentor Information

Barry L. Myers Dale A. Hamilton

Presentation Date

7-2016

Abstract

Supervised classifiers have a wide range of applications from wildland fire ecology to pathology. There are many supervised classifiers that are available, each with their own benefits and drawbacks. For the purpose of this project the k-Nearest Neighbor (kNN) classifier has been utilized with a KD-Tree. There are open source versions of both of these available for use, but by developing our own classifier we are able to optimize both the KD-Tree and the kNN algorithm and to fine tune the accuracy and temporal performance. The project is being used both in the research of prostate cancer detection and classification through Dr. Joe Kronz at Saint Alphonsus and mapping wildland fire extent and severity through NNU’s FireMAP project. Previous iterations of this project were taking around 7 minutes to run on a 4K image, the current version takes a mere 15 seconds.

Comments

Poster #W64

This document is currently not available here.

Share

COinS
 

KD-Tree and k-Nearest Neighbor Image Classification

Supervised classifiers have a wide range of applications from wildland fire ecology to pathology. There are many supervised classifiers that are available, each with their own benefits and drawbacks. For the purpose of this project the k-Nearest Neighbor (kNN) classifier has been utilized with a KD-Tree. There are open source versions of both of these available for use, but by developing our own classifier we are able to optimize both the KD-Tree and the kNN algorithm and to fine tune the accuracy and temporal performance. The project is being used both in the research of prostate cancer detection and classification through Dr. Joe Kronz at Saint Alphonsus and mapping wildland fire extent and severity through NNU’s FireMAP project. Previous iterations of this project were taking around 7 minutes to run on a 4K image, the current version takes a mere 15 seconds.