Face detection using consecutive bootstrapping ICA
TK Kim, CK Choi, Kim · 2002
In this paper, a learning algorithm for frontal face detection based on independent component analysis (ICA) is proposed. An assumption made in this work is that similar patterns to the representative face pattern are faces. The representative face pattern is acquired by averaging various face images with different lighting conditions, shapes, and facial expressions. Input vectors for ICA learning consist of the representative face pattern and other non-face patterns. Non-face patterns are collected with a view to bootstrapping. The bootstrapping is carried out consecutively until there is no more false detection in the non-face images. The multiple steps of bootstrapping and the use of small number of non-face patterns in each step could make the learning process converge. The consecutive bootstrapping steps yield the independent face filters (IFFs). One or more IFFs are obtained at each bootstrapping step. At each step, the probability density functions (pdfs) for the coefficients of the ICA feature vectors are acquired by projecting face images to each ICA feature vector. The feature vectors that are peculiar to their responses to face images qualify as IFFs. Face Detection requires the preprocessing of lighting correction. The non-linear illumination model using a sigmoid function is also proposed. The proposed bootstrapping ICA is novel in that face features are acquired by considering higher order statistics of not only faces but also non-face patterns. Experimental results show that the proposed method outperforms those of previous face detection methods. I.