文本图像中图标图形的识别检索方法

Logo recognition and retrieval algorithm in document image

针对文本图像自动处理中对图标图形识别检索的应用需求,提出一种基于投影特征的图标图形识别检索方法。采用投影直方图表示图标的水平方向特征和垂直方向特征,并利用直方图的相关距离来度量图像的相似性程度,克服镜面对称变换对投影直方图的影响。实验表明,投影特征描述了图标的水平及垂直方向的空间分布信息,能够有效地进行识别检索。

To fulfill the application requirement to retrieve logos in automatic processing document image, a logo recognition and retrieval algorithm based on projective features was presented. The projective histogram was used to denote the horizontal and vertical direction features. To overcome the effects on projective histograms which come from mirror symmetry transform, the similarity was measured by histogram correlation distance in this algorithm. Experimental results show that projective features describe the logo spatial distributing information of horizontal and vertical direction, and can recognize logos effectively.