The face detection has become one of the most important topics in machine learning and computer vision in the last few decades. Many papers have been published utilizing various methods for face detection. One of the most popular face detectors used in many practical applications with heterogeneous computing architecture is Viola-Jones method. Despite of being a real-time and robust face detector, it suffers from not well explained parts at the training procedure, e.g., not clear how to select few features in the first cascades or not clear how many samples are needed and how to gather a good trainset for training a cascade. In this paper, a trainset selection method based on histograms generated from AdaBoost and in addition, simple ways to select few features in beginning cascades are proposed. The training procedure is then compared to a baseline training presented in the previous studies.

在过去的几十年中，人脸检测已成为机器学习和计算机视觉中最重要的主题之一。利用各种方法进行面部检测的许多论文已经发表。Viola-Jones方法是在具有异构计算体系结构的许多实际应用中使用的最受欢迎的面部检测器之一。尽管它是一种实时且强大的人脸检测器，但它在训练过程中的部分内容并没有得到很好的解释，例如，不清楚如何在第一个级联中选择几个特征，或者不清楚需要多少个样本以及如何收集一个样本。训练级联的好火车头。在本文中，提出了一种基于从AdaBoost生成的直方图的火车集选择方法，此外，提出了在开始级联中选择少数特征的简单方法。