视频错帖检测系统及算法设计  

Design of a Video Page Checking System with Recognition Algorithm

作　　者：王跃宗[1] 刘京会[2] 刘喆[1] 

机构地区：[1]北京工业大学机械工程与应用电子技术学院,北京100124 [2]石家庄经济学院公共管理学院,石家庄050031

出　　处：《机械工程学报》2009年第8期233-237,共5页Journal of Mechanical Engineering

基　　金："十一五"国家科技支撑重点资助项目(2006BAF03B01)

摘　　要：为了克服现有光眼检测法和模拟视频检测法的不足,研制一种新的基于互补式金氧半导体工艺(Complementary metal-oxide-semiconductor,CMOS)的图像传感器智能错帖检测系统,其硬件将CMOS图像传感器、数字信号处理器(Digital signal processor,DSP)和可编程逻辑器件(Complex programmable logic device,CPLD)集成在一起,作为图像采集和处理的核心单元,另外还包括通用串行总线(Universal seria lbus,USB)、串口和存储器等辅助电路。检测方法采用反向传播网络(Back propagation network,BP网络)和矩不变量建立错帖识别算法,内嵌在数字信号处理器中,算法适合于文字、图文和图形等多类书帖,检测精度更高。算法参数估计试验表明,相对旋转角在±7o范围内时,检测准确率约为98%。In order to overcome the shortages of current checking methods based on optical eyes or analog video, a novel intelligent page checking system based on image sensors with complementary metal-oxide-semiconductor （CMOS） is designed. In this system, CMOS sensor, chip of digital signal processor and chip of complex programmable logic device are integrated as a core unit for gathering and processing images. And this system also includes auxiliary circuits of USB communication, serial communication and memory. A new algorithm for page checking is also developed on the basis of back propagation neural network and geometric invariant moments, and this algorithm is embeded in DSP chip. Experiments of parameter estimation indicate that this new algorithm is suitable for recognizing different pages, such as word pages, graphic-word pages and graphic pages. Experiments show that recognition precision is about 98% when relative rotation angle is ＋7°.

关 键 词：图像传感器 神经网络 矩不变量 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]