基于张量投票和成份标注的彩色文本的分割  

Text segmentation in color images based on tensor voting and component labeling method

作　　者：魏琳[1] 陈秀宏[1] 

机构地区：[1]江南大学信息工程学院,江苏无锡214133

出　　处：《计算机工程与设计》2009年第2期478-480,共3页Computer Engineering and Design

摘　　要：自然景观中文本会受到各种噪声的污染,这给文本的分割和识别带来困难。根据不同特征值将图像分层,使用张量投票框架对每层进行分析,可以有效地去除噪声并对缺失的图像进行修补。由于要根据经验判断阈值,要达到理想的效果必须经过多次运算。引入经改进成分标注方法,确定阈值,提高了效率。实验结果表明,该方法取得了比较好的效果。In natural scene, text elements are corrupted by many types of noise. There effects make the automatic segmentation very difficult. We identify image using hue and intensity respective and normalization in the same range. Analysis each layers using tensor voting, remove image＇s noise effectively and restore the lost information. Excellent results are achieved by many times compute because confirm the threshold depend on experience. We use improved component labeling method confirm threshold and improve efficiency. Excellent results are achieved in experiments.

关 键 词：文本分割 张量投票 成份标注 中值滤波 阈值确定 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]