摘要
针对复杂背景下文本误检率较高的问题,提出了一种基于蚁群聚类和LBP-HF特征验证的复杂场景文本定位算法。该算法首先利用小波高频系数统计特征表达文本模式,采用蚁群聚类算法对文本像素和背景像素进行分类,得到所有可能的文本区域;然后提取更具区分力的LBP-HF纹理特征对侯选的文本区进行验证,获得文本的准确位置。实验结果表明,所提出的基于LBP-HF特征的验证方法能够有效区分文本和非文本区域,使复杂背景下的文本误检率明显下降。
Aiming at the problem of high error detection rate in complex background text location, a novel algorithm based on ant colony cluster algorithm is proposed for detecting and locating text regions in natural scene images. In the proposed method, firstly, statistical features of wavelet coefficients are used to represent text mode, and then ant colony cluster algorithm is selected to classify image pixels as text regions and background, thus all possible text regions are obtained. Secondly, Local Binary Pattern Histogram Fourier （LBP-HF） feature is applied to remove non-text areas which have the similar characteristics in wavelet domain. Experimental results demonstrate that verification procedure based on LBP-HF is effective, by which the error detection rate is reduced greatly for complex background text detection
出处
《光电工程》
CAS
CSCD
北大核心
2012年第3期137-143,共7页
Opto-Electronic Engineering
基金
国家自然基金资助项目(60872141)
陕西省自然科学基础研究计划资助项目(2009JQ8019)
