文本图像条带污染去除的l_(0)稀疏模型与算法  

l_(0) SPARSE MODEL AND ALGORITHM FOR STRIP POLLUTION REMOVAL OF TEXT IMAGE

作　　者：喻恒 耿则勋 Yu Heng;Geng Zexun(School of Information Engineering,Pingdingshan University,Pingdingshan 467000,Henan,China;School of Geospatial Information,University of Information Engineering,Zhengzhou 450052,Henan,China)

机构地区：[1]平顶山学院信息工程学院,河南平顶山467000 [2]信息工程大学地理空间信息学院,河南郑州450052

出　　处：《计算机应用与软件》2022年第6期183-193,209,共12页Computer Applications and Software

基　　金：河南省科技攻关项目(202102210331)。

摘　　要：针对文本图像受条带噪声污染严重影响视觉效果以及文字正确识别的问题,提出一种基于l_(0)稀疏模型的条带污染文本图像修复算法。该算法在分析条带噪声特性基础上,设计基于l_(0)范数的稀疏正则化模型,实现对条带噪声的有效稀疏约束。通过引入方向相对梯度的自适应正则化参数,使算法不仅适用于规则条带还可应用于适当不规则条带。在准确估计条噪声污染区域的基础上,利用总变分(Total Variation,TV)极小化图像修复作为后处理,形成完整的非凸不可微泛函的优化求解算法,实现规则与不规则条带污染文本图像的良好恢复,弥补了目前方法存在的不足。利用MATLAB软件平台对收集的文本污染图片数据集中的大量图像数据进行实验,与传统条带噪声消除算法进行对照分析,结果表明,该算法修复后的文本图像上中文文字正确识别率提高至少10%以上,英文文字正确识别率提高40%以上,证明了算法的优越性与实用性。The text image is polluted by strip noise,which seriously affects the visual effect and word correctly recognition.To solve this problem,we put forward a stripe polluted text image restoration algorithm based on l_(0) sparse model.After analyzing the characteristics of the stripe noise,this algorithm designed a sparse regularization model based on l_(0) norm to realize the effective sparse constraint of stripe noise.By introducing the adaptive regularization parameter based on the directional relative gradients,the algorithm was suitable for both regular stripe and irregular ones.On the basis of accurately estimating the stripe noise polluted area,TV(total variation)minimization image restoration was used as the post-processing to form a complete optimization algorithm with non convex and non differentiable function,which realized the good restoration of regular and irregular stripe polluted text images,and made up for the shortcomings of the current methods.MATLAB was adopted to conduct experiments on a large number of image data collected in the text pollution image data set.The experimental results show that compared with the traditional strip noise elimination algorithm,the correct recognition rate of Chinese characters and English characters are improved by more than 10%and 40%respectively,which proves the superiority and practicability of the algorithm.

关 键 词：文本图像恢复 条带噪声 l_(0)稀疏模型 自适应正则化 TV图像修复 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]