An Approach to the Feature Selection in the Objects Detection Problem in Images

Workshop on Materials and Engineering in Aeronautics

材料科学;航空航天工程

Demidova, L.A.^1^2 ; Egin, M.M.^1 ; Tishkin, R.V.^3

Ryazan State Radio Engineering University, Ryazan, Russia^1

Moscow Technological Institute, Moscow, Russia^2

Limited Liability Company Areal-98, Moscow, Russia^3

关键词: Classification quality;    Descriptors;    Histograms of oriented gradients;    Neighborhood component analysis;    Objects detection;    SVM classifiers;    Test sets;   

Others  :  https://iopscience.iop.org/article/10.1088/1757-899X/476/1/012007/pdf
DOI  :  10.1088/1757-899X/476/1/012007

来源: IOP

【 摘 要 】

The objects detection problem in images has been considered. It is known that this problem can be solved using the SVM classifier development with application of the descriptors of the histograms of oriented gradients for creation of the training and test sets, but this technics deals with the significant time expenses for the formation of the histograms and the development of the SVM classifier. The application of the special tool known as the feature selection using the neighborhood component analysis algorithm for classification to determine the most informative features of objects has been investigated. It is shown that this approach allow to reduce the time expenses on solving the objects detection problem while maintaining acceptable values of the classification quality indicators. The experimental results on solving the objects detection problem have been presented.