基于Gabor特征与BP神经网络的屏幕显示自动校验系统

摘要

针对人工校验视频监控设备屏幕显示(OSD)效率低下、人力物力资源耗费大的问题,提出一种OSD自动校验系统,取代传统的人工校验方式.系统首先综合多种数理统计特征进行OSD定位,然后利用改进的Otsu算法进行精确字符分割并二值化,最后通过基于Gabor特征离线训练的改进型BP神经网络进行字符识别.实验结果表明,在确保92.7％识别率的前提下,该系统识别一帧OSD平均耗时53 ms.%To deal with low efficiency and long-time consumption in verifying OSD ( On-Screen-Display) of video devices, this paper devised an automatic OSD verification system. The system consisted of three parts. OSD area location was achieved by synthesizing statistical characteristics. Single character was then segmented based on improved Otsu algorithm. Finally, Gabor features and improved BP neural network were used to recognize these characters. The experimental results show that this system costs average 53 ms per recognition of one frame with a recognition rate at 92. 7%.