低质量文档图像中图标的定位

Logo Location in Low-Quality Document Image

针对二值图像提出了基于金字塔模型的目标定位方法 ,该方法适合于构成目标的连通区互相之间距离较小 ,而与其他连通区之间距离较大的情形 首先生成图像的金字塔模型 ,构成目标的连通区会在金字塔模型的某一层合并为一个连通区 ,根据金字塔模型各层中连通区的指示在原始图像中确定一块区域 ,用C4 5作粗分类 ,再用模板匹配判定是否含有目标 把这种方法应用于低质量文档图像中图标的定位 用含 30 0个图标的图标库进行实验 ,得到 88 7%的定位查全率 ,70 4 2

A scheme based on pyramid model is proposed for object location in bi-value images,which is an effective method for such applications as distances among object's composite connected-components,shorter than the distances to others. At first the pyramid model of image is created. Being adjacent to each other,those connected-components belonging to an identical logo will cluster into a single one in certain hierarchy. Then particular regions are determined based on the indication of every connected-component in pyramid hierarchy. Finally template matching is employed to determine if the region includes some objects. This scheme is used to locate logo in low-quality document image. The performance is evaluated by a logo corpus involving 300 logos,the recall of 88.7% and precision of 70.42% are obtained respectively.