一种基于梯度差的文档图像文本行检测算法

摘要

Based on analyzing characteristics of text lines, the algorithm of using horizontal gradient difference was presented to detect text line. Firstly, the horizontal gradient difference was calculated in the gray image, and then the maximum gradient difference was solved in the local window, after text line areas merging, document image shows as inline-blocks. In order to eliminate characters with different heights, the idea of non-text filtering was used in the paper. The result shows that this algorithm has high accuracy as well as good detection rate. In addition, the algorithm still has some robustness and better applicability.%在分析文本行特点的基础上，提出了一种利用水平梯度差进行文档图像的文本行检测算法。该算法首先对输入的文档图像进行水平梯度差计算，然后在局部窗口中求解最大梯度差并进行文本行区域的合并，通过非文本区域过滤来消除字符阶跃的跳变，最后将文档图像以行块的形式进行显示。实验结果表明，与投影算法进行相比，该算法对于行间距较小的文档图像的检测效果较好，时间复杂度较低并且检测的正确率较高，具有一定的鲁棒性和较好的适应性。