Object Detection by Selective Integration of HLAC Mask Features

Higher order local autocorrelation (HLAC) proposed by Otsu [5] is often used in the recent computer vision application such as gate recognition, object tracking, or video surveillance. The feature value of HLAC is the integral of the product of local pixels´ value, and usually the integrals are calculated in entire images. However, in the image recognition, feature selection is often effective for the both of classification accuracy and processing speed. In this paper, we propose HLAC Mask Features extracted from arbitrary local regions, and its feature selection algorithm based on Adaboost technique. We show Adaboost can select HLAC Mask having higher classification power and lower computational cost than usual HLAC for face detection task.