Face detection and eye location using a modified ALISA texture module

This paper presents an automatic method for face detection and eye location using a modified version of the ALISA Texture Module. ALISA (Adaptive Learning Image and Signal Analysis) is an adaptive classification engine based on collective learning systems theory. Using supervised training, the ALISA engine builds a set of multi-dimensional feature histograms that estimate the joint PDF of the feature space for the trained class(es). In the current research, 4 to 6 general-purpose texture and color features are used, which require only a few thousand bins (unique feature vectors) to represent faces for several different ethnic groups by allocating the feature space dynamically. The method first detects the face regions using the ALISA texture module and then locates the eyes inside these regions. A preliminary comparison with a widely-used parametric approach for modeling color information in the presence of changing illumination conditions, demonstrates that the ALISA texture module offers significantly better accuracy for detecting regions of skin. The proposed method also offers competitive speed and is thus feasible for real-time applications to both still images and video sequences.