一种基于聚焦函数的标示牌中文本提取方法

在自然场景下各种标示牌一般背景比较清晰而且边框直线特征明显.针对这一特征,本文提出一种有效的标示牌中文本提取方法.首先利用Roberts算子进行边缘提取,提出基于自适应阈值进行二值图像处理,利用数学形态学中的腐蚀算子消除非直线因素的影响,突出直线特征;然后提出基于聚焦函数的投影方法,找到需要的文本区域,完成文本提取.本算法速度快,定位准确、容易实现,而且对倾斜的直线特征具有很好的兼容性.同时,可以处理各国文字信息,有很好的通用性.

A Method of Text-extracting in Nameplate Based on Focusing Function

Under natural conditions, all kinds of signs commonly have clear background and border featured by the straight lines. Aiming at this feature, this paper proposes an effective text-extracting method. First, the method uses Roberts operator to finish edge extracting, processes binary image based on self-adapting threshold, utilizes erosion operator in mathematical morphology to eliminate nonlinear influence and outstand linear feature. Then, the method puts forward focusing function based projection, finding text region required and completing text extraction. The method has high efficiency, locates precisely and easily implement. Moreover, it has good compatibility for tilting linear feature. In the meantime, it can process text information in difficult languages with its commonality.