A Fast and Accurate Face Detector
Based on Neural Networks
´raud, Olivier J. Bernier, Jean-Emmanuel Viallet, and Michel Collobert
—Detecting faces in images with complex backgrounds is a difficult task. Our approach, which obtains state of the art results,
is based on a new neural network model: the Constrained Generative Model (CGM). Generative, since the goal of the learning process
is to evaluate the probability that the model has generated the input data, and constrained since some counterexamples are used to
increase the quality of the estimation performed by the model. To detect side view faces and to decrease the number of false alarms, a
conditional mixture of networks is used. To decrease the computational time cost, a fast search algorithm is proposed. The level of
performance reached, in terms of detection accuracy and processing time, allows to apply this detector to a real world application: the
indexation of images and videos.
—Combination of models, face detection, generative models, machine learning, neural networks, projection.
detect a face in an image means to find its position in
the image plane (x,y) and its size or scale (z). Two broad
classes of algorithms can perform this task.
An image of a face can be considered as a set of features
such as eyes, mouth, nose with constrained positions and
size within an oval: an explicit model can be used. One of
the simplest and fastest methods to realize the feature
extraction is the projection of the image or the edge image
on the vertical axis to find the eyes or the mouth and on the
horizontal axis to locate the nose , , . Several other
methods are currently used to perform the feature extrac-
tion: Gabor filter , oval detection , , etc. A
similarity measurement between features is then used for
face recognition or face detection task: Mahalanobis dis-
tance , crosscorrelation , , , graph matching ,
elastic matching of features , decision tree , neural
network , belief network .
Considering that an image of face is a particular event in
the set all the possible images, extracted windows of the
image can be analyzed to determine if these windows
contain faces or parts of background. A probabilistic or
statistic model can be used to analyze the pixels intensity of
each subwindow (extracted window of the image). This
model can be built with different methods: neural networks
, , , , , , , , , , principal
components analysis , , , , , , Kullback
distance and maximum-likelihood method , Support
Vector Machines , , etc.
For face detection, the advantage of explicit models is
usually the speed of the features extraction algorithm and
the similarity measurement task in comparison to the
methods directly based on the analysis of pixels intensity.