Abstract: An adaptive fast image enhancement algorithm is developed to improve the face detection under complex illumination environment.The proposed method processes face images in two separate steps：Dynamic range compression and detail enhancement.Dynamic range compression is logarithmic and non-linear function transformations which are not only able to enhance the luminance in dark regions，but also effectively inhibit high-light regions.Then unsharp mask filtering is utilized to enhance the image detail.With variety of enhancement techniques and Adaboost face detection algorithm，comparative experiment is made in Yale B face database.Experiment results demonstrate that the proposed image enhancement algorithm can efficiently improve face detection rate and reduce false detection rate，and also has better performance compared to histogram equalization，single-scale Retinex and multi-scale Retinex.

Key words: face detection, Adaboost, self-adaptive image enhancement, histogram equalization, single-scale Retinex, multi-scale Retinex

摘要： 提出了一种自适应的快速图像增强算法用于改善复杂光照下的人脸检测。算法对人脸图像的增强分为两步：动态范围压缩和细节增强。算法首先利用对数变换和非线性变换，增强图像阴暗区域的信息，同时对高光区域进行有效地抑制，然后利用反锐化掩模滤波对图像的细节进行增强。将各种增强算法应用于图像的预处理，结合Adaboost人脸检测算法，在Yale B人脸数据库上进行对比实验。实验结果表明，自适应快速图像增强算法能有效提高人脸检测率和降低误检率，具有比直方图均衡算法、单尺度Retinex算法和多尺度Retinex算法更好的性能。

关键词: 人脸检测, Adaboost算法, 自适应图像增强, 直方图均衡, 单尺度Retinex, 多尺度Retinex