摘要
针对移动终端处理能力低、内存小等影响系统效率的问题,提出了三阶段视频字符实时识别方法:视频采集及图像预处理、字符区域定位和字符识别。对于字符区域定位,提出了基于感兴趣区域(ROI,Region of Interesting)运动检测的相似帧过滤算法,并通过数学形态学与连通区域相结合的方法进行字符定位;对于字符识别,提出了基于误差阈值筛选的多模板字符识别算法,保证较高识别率。算法均采用NDK开发框架实现。实验结果表明,该方法在每个阶段都提高了处理效率,达到了对视频字符实时识别的效果。
To solve the problem of mobile terminals for low processing capacity, small memory and otherfactors which have effects on system efficiency, a 3 stage method of real-time video character recognitionbased on Android platform is proposed. The 3 stages are： video capture and image pre-processing,character regional positioning, character recognition. For character regional positioning, an algorithm ofsimilar motion detection frame filtering based on ROI motion detection is presented, and the characterlocating algorithm mainly uses the combination of mathematical morphology and connected regionanalysis. The algorithm ensures high accuracy and processing efficiency. Then the multi-templatecharacter recognition algorithm based on error threshold is proposed for the character recognition stage,and the recognition rate is improved. All algorithms are based on the NDK development framework toimprove processing efficiency. The experimental results show that, 3-stage method improves processingefficiency at each stage, thus improving the overall efficiency of the mobile platform video characterrecognition speed reached on the Android platform real-time video character recognition results.
出处
《信息技术》
2014年第7期37-41,共5页
Information Technology
基金
国家自然科学基金(61103142)
