Determining Effective Features for Face Detection Using a Hybrid Feature Approach

Abstract

Detecting faces in cluttered backgrounds and real world has remained as an unsolved problem yet. In this paper, by using composition of some kind of independent features and one of the most common appearance based approaches, and multilayered perceptron (MLP) neural networks, not only some questions have been answered, but also the designed system achieved better performance rather than the previously presented works. The designed face detection algorithm is composed of two main stages; in the first stage of the algorithm, color based skin detection is performed using some selected color space components from the whole 15 color space components in order to reduce the search space and in the second one, verification of detected regions is done using some other kinds of different features including texture, gradient, image and geometric features. Unlike the other studied issues, in this paper, various types of features aren't evaluated in separated algorithms and systems; rather they compete all together in one competitive learning vector and after the training of the neural network the system participates in the process of feature selection. Using designed method and besides of dimensional reduction of input matrix extraordinarily, each chosen feature was ranked.