一种面向聚焦爬虫的自然场景文本定位技术

Locating Natural Scenes Text for Focused Crawler

各类视频设备采集的图像中包含了高质量的海量主题信息.尽可能全面地找到自然场景中的文本,对聚焦爬虫具有重要意义.提出一种具有较高召回率的文本定位技术,充分注意到角点强度极大、极小值对自然场景文本定位的重要作用,运用提出的角点强度增强技术和三值化方法,能够有效地分离出极大、极小值区域,从而,能够将图片中的大部分文字从复杂的背景和各种噪声中凸显出来.实验表明,该方法能够大幅提高基于角点特征的定位算法的召回率,为聚焦爬虫从大量的自然场景图片中获取重要的主题信息提供了可能.

Vast amounts of natural scene pictures contain subject information with high quality, and it＇ s a new subject information resource for focused crawler. We propose a novel comer analysis technique for natural scene text location. Considering both the minimum and the maximum of comer intensity clearly remark the locations of text in natural scene, we non-lineally enhance the comer intensity of every pixel, and then binarize it. As we show, the binarized comer map shows the salient text blocks. The experiment result shows that our comer analysis technique can help the text localization algorithm based on its significantly improving its recall. Our corner analysis technique can help focused crawler crawl more real-time subject information from vast amounts of natural scene pictures captured by mobile and web camera.