The best way to conference proceedings by Francis Academic Press

Web of Proceedings - Francis Academic Press
Web of Proceedings - Francis Academic Press

Image Classification Algorithm Based on Depth Neural Network and Multi-Level Feature Learning

Download as PDF

DOI: 10.25236/csam.2019.067

Author(s)

Qian Sun, Li Xin, Faliang Chang, Zengshun Zhao

Corresponding Author

Qian Sun

Abstract

Image classification being widely applied to computer vision is an im-age processing method to distinguish the different category targets according to the different features reflected by the image information. BOW-SVM is a relative-ly typical image classification method with higher precision, however, it’s unsatis-factory in operation performance. To improve the performance and precision more efficiently, a high-efficiency image classification method based on HOG-PCA is proposed. First of all, it is to make the feature whitening by extracting the Histogram of Oriented Gradients (HOG) features, secondly, make the random down-sampling for the scale unification, afterwards, adopt the principal component analysis (PCA) for feature mapping and finally make the nearest neighbor classification through the minimum two-order norm determination. In the experiment, the proposed method is realized and tested on the P ASCAL 2012 data set through C++ on the basis of OPENCV and Darwin to compare the precision and operation performance of this method and BOW-SVM method; according to the experiment, the proposed has higher precision and better operation performance.

Keywords

Deep Neural Network, Multi-Layer Feature Learningl, Minimum Two-Order Norm, Image Classification, Image Features