一种抗噪的自然场景图片预处理算法模型

Denoising of Natural Scene Image Preprocessing Algorithm Model

随着大数据时代的到来,基于自然场景图片的文字识别技术将大大提高海量图像内容检索的效率,为了利用成熟的OCR软件将自然场景中的文本识别出来,提出一种抗噪的自然场景图片预处理算法模型.该算法模型分为文本区域筛选、文本区域二值分割和文本校正三步骤,针对这三步骤分别提出了对应的三个算法.文本区域筛选采用基于角点的笔画宽度转化算法,文本区域二值化分割采用一维Otsu的双斜率分割法,文本校正采用基于Radon变换的分步投影算法.通过四个实验的验证结果可知,该算法模型文本区筛选精度高、抗噪性能强、算法复杂度适中,能适合各种角度的文本区域识别.

With the arrival of the era of big data, pictures of natural scenes of character recognition technology will greatly improve the efficiency of massive image content retrieval, to take advantage of mature OCR software to text recognition in natural scene, this paper proposes a noise model of the natural scene image preprocessing algorithm. The algorithm model is divided into the text area selection, text areas binary segmentation and text correction three steps,for the three steps are corresponding to the three algorithms are put for- ward. The text area screening uses the stroke width conversion algorithm based on comer point, the text area binarization segmentation uses one-dimensional dual slope of Otsu segmentation method, and the text correction uses the step by step based on Radon transform projection algorithm. Through the results of four experiments shows that the algorithm model text area screening of high precision, strong antinoise performance,the algorithm complexity is moderate,and fit the text area of all kinds of Angle recognition.