通信工程文献翻译——本文提出了一种基于 CostSensitive A daBoost（CS- A daBoost）算法在图像平面中检测任意旋转角度的新颖方法。 该方法首先使用由 CS-A daBoost训练的非常简单的分类器级联来确定每个输入窗口的可能取向，然后使用也由 CS- A daBoost 训练的直立面部检测器来验证被解旋转的面部候选物。 本文还给出了两种分类器的训练程序。 实验结果表明，与传统方法相比，所提出的方法具有更高的检测率和实时检测速度。

1介绍

在这几年中，由于人脸识别，人机交互等领域的广泛应用，人脸检测的问题引起了广泛的关注。许多技术已经应用于问题，如概率估计[1]，神经网络[2 ]和 AdaBoost [3]。 虽然这些提出的方法在检测直立正面时具有良好的性能，但是定位旋转面仍然是一个困难的问题。

为了在图像平面中以任意程度的旋转方式将直立面的方法扩展到面部，可以尝试两种方法：一种是将图像一步一步地旋转 360度，然后使用直立面检测器以检测每个旋转的图像中的面。然而，这将是一个计算上非常费时的任务。即使现在的实时方法，它可以检测每秒15帧的立面，它们只能以每秒 1?2帧的速度检测旋转的面。另一种方法是首先确定面部候选者的可能的旋转角度，然后使用直立的前脸检测器来进一步验证旋转的候选者。应该提到

的是，不包含脸部的窗口在旋转后不会包含任何脸部。在[4]中，系统使用“路由器”神经网络进行分析输入窗口，并在“检测器”网络处理之前返回可能的结果。在方法[5] [6]中，使用方向直方图或 Hausdroff 距离来确定面部候选者的可能方向。虽然这些方法可以实现良好的检测性能，但所有这些方法在计算上都相当费时，并不适合实时应用。与以前的系统不同，本文提出了一种完全基于统计学习算法的高效旋转面部检测方法。我们的方法首先使用由 CS- A daBoost 学习算法训练的简单分类器级联来确定可能的面部取向，然后使用直立面部检测器进一步验证解旋窗。 因为面部取向分类器比直立面部检测器简单得多，并且直立检测器仅在每个图像位置应用一次，所以我们的方法要比彻底地尝试......

ABSTRACT 

This paper presents a novel method of detecting faces at any degree of rotation in the image plane based on CostSensitive AdaBoost (CS-AdaBoost) algorithm. The method first employs a cascade of very simple classifiers trained by CS-AdaBoost to determine the possible orientation of each input window and then uses an upright face detector also trained by CS-AdaBoost to verify the derotated face candidate. The training procedures for both classifiers are also given in the paper. Experimental results show that the proposed method gives higher detection ratio and real-time detection speed compared to the conventional ones. 

1. INTRODUCTION 

In these years, the problem of face detection has attracted much attention due to its wide applications in face recognition, human-computer interaction, etc. Many techniques have been applied to the problem, such as probabilistic estimation [1], neural networks [2] and AdaBoost [3]. Although these proposed methods have good performance in detection of upright frontal faces, locating rotated face is still a difficult problem. To extend the methods for upright faces to faces at any degree of rotation in the image plane, two ways can be tried: One is to rotate the image by a small degree, step by step up to 360 degree, then use the upright face detector to detect faces in each of the rotated images. 

However, it would be a computationally expensive task. Even for the current real-time methods [3][9], which can detect upright faces 15 frames per second, they can only detect rotated face about 1~2 frames per second. The other way is to first decide the possible rotation angle of the face candidate, and then use upright frontal face detector to verify the derotated candidate further. It

should be mentioned that a window that does not contain......