基于HOG+SVM模型的场景文字二次检测算法

Scene Text Secondary Location Algorithm Based on HOG+SVM Mode

针对基于边缘检测的文字定位虚警率过高的问题,提出了一种基于Canny边缘检测和HOG+SVM模型相结合的场景文字检测算法。首先采用基于Canny边缘检测和文字的几何约束条件得到候选文字区域,再利用HOG+SVM模型对候选文字区域进行二次检测,过滤掉大部分非文字区域。实验结果表明,该算法能够有效地去除基于边缘检测算法产生的误检区域,大大降低了基于边缘检测的场景文字定位的虚警率,并对背景复杂的图像也具有一定的鲁棒性。

To reduce the false alarm rate in the scene text location algorithm based on edge detection, a scene text detection meth- od based on the combination of Canny edge detection and HOG ＋ SVM mode is proposed. Firstly, the candidate text regions are ex- tracted by the combination of Canny edge detection and the word＇ s Geometric constraints. Secondly, most of the non - word candi- date text regions are deleted by HOG ＋ SVM mode. Experimental results show that the proposed method can filter out the false de- tected regions, whereas the false alarm rate based on the edge detection is reduced greatly. In addition, the proposed method can also deal with the complex scene images well.