关键词: 矩不变量, 神经网络, 图像传感器

Abstract: 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 ±7o.

Key words: Image sensors, Invariant moments, Neural network