摘要： 提出了一种使用改进的AdaBoost 分类器来检测体育场景的方法。将电视新闻中的体育场景分为三类：草地运动，冰雪运动和人造场地运动。针对这几种不同的体育场景，提取颜色直方图、边缘方向直方图和共生矩阵纹理等3 种低层视觉特征，然后用改进的可自动选择特征的boosting 方法为每一类体育场景分别建立AdaBoost 分类器。该文提出的方法应用在国际视频处理评测TRECVID2003 中的“体育场景”语义特征抽取任务上，取得了很好的效果。

关键词: Boosting；弱分类器；体育场景检测

Abstract: This paper proposes a method for detecting sports scenes using variant AdaBoost classifier. In the approach, sports scenes in TV news are initially classified into three types: on grass, ice or snow and man-made floors. Three low-level image features of color histogram, edge direction histogram and co-occurrence matrix texture are extracted. The focus of this paper is on the use of the boosting method for automatic selection of features and classification. For each type of sports scenes, it builds a classifier based on AdaBoost using low-level image features above. The category confidence scores of each image calculated by each classifier are then combined into the final result. The paper applies the system on the extraction task of semantic feature “Sports Event” of TRECVID2003. It works effectively and gives high precision.

Key words: Boosting; Weak classifier; Sport scene detection