The Recognition and Classification of Objects Based on the Modified Distance Metric

Abstract

The recognition of objects in image based on a specific pattern and distance metrics allows solving the tasks of classification and object search. The methods for recognition, classification and search of the objects are considered. Methods are used to recognize faces in images using various distance metrics. In recognition methods, the accuracy of the result depends on the correct choice of the pattern and the distance metric. In the methods of image recognition, the following distance metrics are used: Euclidean distance, squared Euclidean distance, normalized squared Euclidean distance, angular cosine distance, correlation coefficient distance, mean Euclidean distance, mean squared Euclidean distance, mean squared root distance. Using one of the metrics in the recognition methods does not always give an accurate result. For the considered problem of face recognition, the following distance metrics are applied: Euclidean distance, squared Euclidean distance, normalized squared Euclidean distance, angular cosine distance. To improve the quality of object recognition, a new generalized distance metric is proposed. The mathematical model of the new generalized distance metric is presented in the work. The application of the new generalized metric in the problem of face recognition showed a more accurate result in comparison with standard metrics and improved recognition by 20%.