CAJ | 학술논문

针对光照不均文本图像识别率低下的问题，提出一种基于区域分类的自适应校正算法。通过八邻域三维增强，拉大背景点与目标点差异；通过对该类文本图像的特征分析，利用投影算法实现图像的区域分类；采用全局校正处理区域分类后的页白区域和均匀区域，结合评定图像质量的特征参数，对阴影区进行局部校正处理，实现背景与目标的彻底分离；分析噪音分布特点，采用邻域去噪法纠正误判像素点。实验结果表明，该算法能有效分离光照不均文本图像的目标点和背景，处理后的效果图可用于阅读以及提高OCR系统识别率。
침대광조불균문본도상식별솔저하적문제，제출일충기우구역분류적자괄응교정산법。통과팔린역삼유증강，랍대배경점여목표점차이；통과대해류문본도상적특정분석，이용투영산법실현도상적구역분류；채용전국교정처리구역분류후적혈백구역화균균구역，결합평정도상질량적특정삼수，대음영구진행국부교정처리，실현배경여목표적철저분리；분석조음분포특점，채용린역거조법규정오판상소점。실험결과표명，해산법능유효분리광조불균문본도상적목표점화배경，처리후적효과도가용우열독이급제고OCR계통식별솔。
To deal with the low recognition rate of the uneven illumination of text image ,an adaptive correction algorithm based on regions was put forward .Firstly ,the difference between the background points and destination points was pulled through the eight fields of three‐dimensional enhancement .Then ,through the analysis of characteristics of the text image ,the image’s re‐gion was classified by using the projection technique .The global correction was used to deal with the non‐text area and the uni‐form illumination area ,and combining with characteristics parameters for evaluating image quality assessment ,the shaded area was corrected locally to achieve complete separation of backgrounds and objectives .Finally ,the characteristic of image noise dis‐tribution was analyzed ,and the neighborhood denoising method was adopted to correct mistakes pixels .The experimental results show that this algorithm can effectively separate the target pixel and the background pixel of uneven illumination of text image , and the visual rendering after processing can be used for reading and improving the recognition rate of OCR system .