Object detection with a minimal set of examples using convolutional PCA

Object and face detection is a very active research topic. Most of the current object detection systems use machine learning techniques like Gaussian Mixture Models [1], Support Vector Machines [2], Adaboost [3] or Neural Networks [4][5]. Some of them use a feature extractor in order to reduce the dimension of the face image space. In [1], face detection is performed using GMMs to extract face descriptors and a Multilayer Perceptron (MLP) to perform classification. In [3], the system performs fast object and face detection using Haar functions and machine learning based on the Adaboost method. Other methods use classifiers directly on image pixels. In [4], Rowley et al. use partially connected Multilayer Perceptrons in order to focus on the local information of the images and to reduce the number of connections of the MLP. In [5], Garcia and Delakis perform face detection using Convolutional Neural Networks. CNNs are powerful bioin-spired hierarchical multilayer neural networks which apply a cascade of learnt convolutional and supsambling filters. The initialization of these systems needs a large training database of object images, manually annotated, which requires time-consuming and costly efforts.