This paper is just a part of the face recognition project which is being developed in Tomsk Polytechnic University. The main goal of that project is to build a system of face recognition in the video stream in real time. Several well-known methods of image processing such as filtering in the frequency domain, difference of Gaussians, Haar wavelet transform, Gabor filtering, log-Gabor filtering and feature extraction such as standard deviation, discrete cosine transform, Hu moments, histogram of oriented gradients that can be used for solving the problem of human face identification are described in this paper. Some ways of applying their combinations in feature vector extraction algorithms are presented. The library of computer vision OpenCV was used in our study. The effectiveness of the proposed algorithms was tested using Caltech Faces base. The results are showed in the comparison chart. The effectiveness comparison was based on the equal error rate, calculated for the false accept rate and false reject rate. The conclusion about the suitability of the most efficient algorithm for the identification problem summarizes this paper. Algorithm that consists of three steps: finding difference of Gaussians, log-Gabor filtering and standard deviation calculation found as the most efficient during this study.
|Состояние||Опубликовано - 2016|
ASJC Scopus subject areas
- Computer Vision and Pattern Recognition