Abstract
This study presents an analysis of the causes of insufficient efficiency of the nearest neighbor method, compared with deep learning networks. The primary cause is the incorrect use of Euclidean distance to the nearest neighbor for estimating the distance from the analyzed pattern to a region occupied by a class. To overcome this problem, it is necessary to construct a local estimate of the distance metric. Therefore, the proposed method combines the concepts of “tangential distance” and Mahalanobis distance. Based on this method, a modified nearest neighbor method is suggested. Validation experiments on MNIST data show that the new method reduces the recognition error rate from 3.8% to 0.8%, which is lower than the recognition error rate of 1.1% for the nearest neighbor method modified based on the concept of “tangential distance,” the calculation of which requires a priori information on allowable pattern transformations.
© 2017 Optical Society of America
PDF Article
More Like This
Cited By
You do not have subscription access to this journal. Cited by links are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.
Contact your librarian or system administrator
or
Login to access Optica Member Subscription