新闻图像中重要人物的自动检测和识别研究

针对网络新闻图像中人物的不同表情、光照、姿势等因素带来的视觉空间差异造成的人脸检测效果差的问题，提出滑动窗口自适应方法和肤色特征相融合，检测新闻图像中的人脸。针对单样本条件下的人脸识别问题，提出有监督的双向二维局部保持投影法，提取人脸特征。结合图像子特征加权方法，先将人脸图像分块，子图像的子集作为一个子特征统计权重，采用人为投票的方式进行识别。实验选取雅虎新闻图像和新华网新闻图像。结果表明，滑动窗口自适应算法在不同类别新闻图像中人脸检测率提高了3%到10%，人脸识别率在一定范围内提高了5%。

Automatic Detection and Identification of Important Persons from News Image Collections

In order to solve the problem of poor face detection caused by visual space differences with different expression, illumination, posture and other factors in the network news images, the method of Adaptive Boosting Window combined with the skin color feature is proposed to detect faces in the images of the news. For the single sample face recognition problems, we present a supervised two-way with two-dimensional locality preserving projection method to extract the facial features. Combined with sub-feature weighting of the image, firstly, we make face images into blocks, each subset of blocks as a sub-feature and statistics weights of each sub-feature, finally, identify by voting. The experiment selected Yahoo and Xinhua news images. The results show that the presented method can increase the face detection rate between 3% and 10% and the face recognition rate increased 5% in some degree.