基于LBP和FSVM的视频文字定位方法

摘要

This paper proposes a video text localization method based on Local Binary Pattern(LBP) and Fuzzy Support Vector Machine(FSVM). The edge detection and morphological operations are used to the video images. The candidate text regions are detected using project analysis and heuristic rules. The final results are defined using FSVM, in which LBP that describes the local edge distribution are used as input features. Experimental results show that the performance of this method is well and it has stronger robustness.%提出一种基于局部二值模式(LBP)和模糊支持向量机(FSVM)的视频文字定位方法.利用边缘信息和形态学操作进行文字粗检测,采用投影直方图和启发式规则形成候选文字区域,提取LBP作为纹理特征,用FSVM对候选文字区域进行精确定位,生成最终的文字块.实验结果表明,该方法具有较好的视频文字定位能力且鲁棒性较强.