基于形态学和窗口特征的快速文本提取

Fast text extraction and its application based on morphology and window features

摘要:

提出了一种快速有效提取目标文本的方法。首先用边缘检测和数学形态学方法粗定位出所有文本区域和图形区域，然后通过多个窗口特征筛选出文本区域，最后设置判别规则提取出目标文本。由于减弱了背景区域的干扰，该算法具有较高定位准确度和较低时间复杂度。同时二值化时使用分块OTSU算法，增强了对复杂背景图像的适应性。通过对多类样本的测试，表明该算法具有较强的通用性，而较高的定位准确率证明了该算法的有效性。

Abstract:

This paper presents a fast and efficient way to extract the target text areas. First, the edge detection and mathematical morphology are used to realize coarse location of text and graphic areas. Then several window features are extracted to distinguish the text areas from graphic areas and thereafter certain rules are set to extract the target text areas. The algorithm has high location accuracy and low time complexity due to weakening the interference of background areas. At the same time, OSTU algorithm is applied to enhance the adaptability of images with complex background during binarization. Test results on several types of samples indicate that the algorithm has strong popularity. The high accuracy of the method has also proved the validity of the algorithm.