过渡映射耦合改进的阈值分割的文本提取方法

Text extraction method based on transition mapping coupled improved threshold segmentation

针对文本提取技术难以准确定位文本区域的问题,提出一种场景文本检测与提取方法。根据文本与其相邻背景之间的瞬态颜色差异,基于像素强度的变化,构建过渡映射,生成一个过渡图;通过计算过渡像素与其周围纹理的一致性,确定候选文本区域;利用LBP算子计算过渡像素附近的强度变化,得到文本区域,利用像素投影优化文本区域,精确定位文本区域的边界;在过渡像素中添加一个约束,利用改进的阈值分割方法,从文本区域中准确提取文本字符串。实验结果表明,与当前场景文本提取技术相比,在复杂视频场景中,所提算法具有更高的文本提取精度与鲁棒性。

To solve the defects of inaccurate localization of text regions in current video text extraction technology,a method of scene text detection and extraction was proposed.According to the transient color difference between the text and its adjacent background,a transition map was constructed based on the change of pixel intensity to generate a transition map.The candidate text region was determined by calculating the consistency between the transition pixel and its surrounding texture.The LBP was used to calculate the intensity change near the transition pixels to obtain text region,and the pixel region was optimized using pixel projection in the transition map so that the boundaries of the text region were accurately located.The text strings were extracted from the text region based on the improved threshold segmentation method and adding a constraint to the transition pixels.Experimental results show that the proposed algorithm has higher text extraction accuracy and robustness in complex video scene compared with other video text extraction algorithms.