Face/Non-face Classification Method Based on Partial Face Classifier Using LDA and MLP

In this paper, we proposed a face / non-face classification method based on a partial face classifier using LDA and MLP for face detection system. General classifier for face and non-face classification requires many face images and background images for training. In contrast, the classification method based on the partial face classifier uses a constraint of facial structure that a pattern on the center of face must be surrounded with patterns on the boundary of face. So the partial face classifier in this method can reduce the affect of non-face image because it´s trained by only face images without non-face images. For improving the classification performance and computation efficiency of partial face classifier, the input pattern has to be represented as a low dimensionality by removing redundant and irrelevant features in the high dimensional image. To achieve this purpose, we transform input image into low dimensional pattern by LDA. In the experimental results, the time and space complexity are reduced by 48.9% and 43.6% when the 8 LDA features are used instead of image features. Also in the experiment of partial face classification, the proposed method which uses 8 LDA feature obtains the classification rate 1.7% higher than using image features.