CAJ | 학술논문

图像中文字信息的提取和识别具有极大的应用价值，文字定位是文字提取最关键的步骤之一。现有文字定位算法复杂度高，查全率和查准率都有待提高。本文提出一种基于频域纹理特征的图像文字定位方法。在图像频域提取纹理特征后初步定位文字区域，再根据文字区域的几何属性得到最后的定位结果。实验结果证明，该方法在粗定位和精确定位阶段均能取得较好的效果，且计算复杂度较低。
도상중문자신식적제취화식별구유겁대적응용개치，문자정위시문자제취최관건적보취지일。현유문자정위산법복잡도고，사전솔화사준솔도유대제고。본문제출일충기우빈역문리특정적도상문자정위방법。재도상빈역제취문리특정후초보정위문자구역，재근거문자구역적궤하속성득도최후적정위결과。실험결과증명，해방법재조정위화정학정위계단균능취득교호적효과，차계산복잡도교저。
Text extraction and recognition in image has great application value,text detection is the cru-cial step of text extraction.Current text location algorithms have high complexity,precision and recall need to be improved.This paper presents a universal approach for text location based on frequency do-main textures.The new algorithm extracts textures feature form the frequency domain,then determine the final text regional on the basis of the geometrical properties of the text.The experimental results demonstrate the low complexity of the proposed algorithm,which can achieve satisfactory results both in the stage of Preliminary detection and accurate detection.