摘要： 提取文字信息是基于内容的视频检索中重要内容。该文综合应用小波变换技术和颜色聚类技术提取含有中文字幕的视频图像的文字，并利用印刷中文字体特点进行噪音处理得到完整清晰的字幕文字。在进行颜色聚类时，该文提出一种8邻域颜色聚类方法，充分利用图像像素的空间相关性，使聚类具有较好的鲁棒性。实验结果表明这是一个有效的中文字幕文字的提取方法。

关键词: 小波变换, 颜色聚类, 中文字幕, 基于内容的视频检索

Abstract: Text information retrieval is an important field in content-based retrieval of video. This article integrats wavelet transform and color clustering to retrieve Chinese caption in video images and performs denoising according to the characteristics of print Chinese letters. While performing color clustering, it proposes an 8 neighbors clustering method which utilizes spatial correlation to achieve a good robust performance. The experiments demonstrate that it is an effective method to retrieve the Chinese caption.

中图分类号: 

• TP391.41