The Research Of Optical Character Recognition Orient Digital Resource Aggregation Platform

Mobile equipment based on Android or i OS system have been widely used in the last decades as long as the internet is being more and more popular. The demand of reading on mobile terminals has also become higher than before. And the construction of digital resource aggregation platform is also in full swing. But it is difficult to make paper resources and digital resources cooperate with each other because they are in different kinds. So an application that can overcome the harsh environment of digital resources aggregation platform and adapt to users’ demands is very important.The concept of OCR(Optical Character Recognition) was suggested in late 1920 s and it has been developed for many years. OCR is a simple and trivial quest for human beings, but it is very difficult to let an app to do so. Because of the restrictions on technology, there are lots of variable factor, such as cameras and environment, which have a heavy impact on the recognition. Most of the traditional methods of OCR are based on statistics, such as template matching. But OCR has been more and more popular in artificial intelligence and machine learning in pace with the developing of technology and theory. And OCR has also been lots of professors’ research area. So this article it is meaningful to find a way to combine OCR and machine learning.There is also a rapid developing area on machine learning in late years which is called deep learning. This concept comes of the research of artificial neural networks. Deep learning can overcome the high complexity when training networks, abstract meaningful information and wipe off redundant or useless information. Neural network usually plays the role of classifier in an OCR system, and deep learning does so of course. This article combines the highly abstract ability of deep network and tuning ability of shallow network. And with the help of tests, we confirm the parameter of the network, quantify influence of image pre-processing and robustness of the system. All the tests shows us that Deep & Shallow Network has advantage over other algorithms.Although Deep & Shallow Network has advantage, it is not perfect. The system has already can complete basic quests through the application. We will keep on mining the potential of deep learning and raise the precision of recognition. We will also find different ways to increase robustness and decrease complexity.