基于NSCT图像文字信息提取新方法  

New method of text extraction from image based on nonsubsampled contourlet transform

作　　者：种耀华[1] 张久文[1] 董敏[1] 

机构地区：[1]兰州大学信息科学与工程学院,兰州730000

出　　处：《计算机应用》2012年第A02期182-185,共4页journal of Computer Applications

摘　　要：采用了非下采样Contourlet变换(NSCT)与动态阈值二值化相结合的方法对图像内的文字信息进行提取。首先对图像进行空间多分辨率变换,再利用NSCT得到图像的高频子带信息和低频子带信息。因高频子带含有丰富的纹理细节信息,而低频子带则含有图像的概貌信息,故将低频子带去除,再结合K-means算法对高频子带进行分类和能量变换,并选取动态阈值对子带进行二值化处理,筛除背景信息。最后综合各分辨率图像的实验结果进行定位,得到了文字区域。实验结果表明该方法能够准确捕获图像文字区域,提取得到理想的图像文字信息,并且具有对文字大小以及噪声鲁棒性好,对图像通用性强的优点。A new method was proposed to extract the text information from the image which combines NonSubsampled Contourlet Transform （NSCT） with the binarization of dynamic threshold. Firstly, the image was transformed into multi- direction subbands by using the NSCT, which could get high and low frequency information of the image. The high frequency subbands contain abundant textual and detail information of the image. Combined the K-means algorithm and the energy of high frequency subbands, high frequency subbands were categorized into strong and weak bands, and the weak bands were boosted. Then, the binarization was applied to the subbands with dynamic threshold so as to filter out the background. Finally, combined with the handled results of all sub-images, the text area was obtained. The results show that this method has obvious advantage which could extract the ideal text information and is insensitive to font-size and noise for most kinds of images.

关 键 词：非下采样CONTOURLET变换 文字定位 文字提取 动态阈值 K—means 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]