DCT and HOG Feature Sets Combined with BPNN for Efficient Face Classification
Bouchra Nassih, Aouatif Amine, Mohammed Ngadi, Nabil Hmina · Procedia Computer Science · 2019
The problem of face classification is to classify faces based on their visual appearance of the faces. In this paper, we investigate this issue by means Back-Propagation Neural Network (BPNN) combined with both descriptors, which are the Discrete Cosine Transform (DCT) and the Histograms of Oriented Gradient (HOG). In the first case, the BPNN is used individually for feature extraction and classification, in the second case, the DCT and HOG descriptors are used to extract features of each image, then, these extracted features are utilized as the input of BPNN to classify faces and non-faces. We perform an experimental evaluation of two facial databases, BOSS, and MIT. Our proposed approach which is (HOG, BPNN) has a significant improvement on classification rate compared to other used methods getting 99.4% and 98.03% in terms of accuracy on BOSS and MIT databases respectively.