Research And Implementation Of Website Code Recognition Based On RPROP Artificial Neural Network

As the development of Internet technology, issues of Internet security are gradually evident. The generation of website code is to prevent insecure factors of Internet like malicious code breaking, repeatedly vote, spam and so on. Website code is an open automatic program for compelling human interact with computer to distinguish them, it is now universally applied to improve web site's security and ability of against attack.Recognition of website code is to enable computer to recognize the characters in the code image which is sent to monitor by sever, by technologies like Image Processing and Artificial Intelligence. Although it violates Internet's security principle, generation and recognition of website code promote each other, the former compels the latter to develop, and new generation technology is a challenge to Image Processing and Artificial Intelligence. In this paper, several image denoising and image segmentation algorithm is brought forward according to characteristics of website code image. Implementation of the RPROP ANN class for website code recognition and how to apply it are presented, for reducing the training time, a new method that to accelerate ANN training by GPU with OpenCL support is brought forward.This work is mainly reflected in the following aspects:Choose character-type website codes with various characteristics, introduce image processing like grayscale and binarization, then improve binarization processing based on actual demands, and design image denoising and image segmentation algorithm as well, as a preparation of the training of supervised learning ANN.Expatiate the RPROP ANN algorithm, design a ROROP ANN class accord to the demand of website code recognition, and present how to apply it to carry out the work, then introduce the website code recognition software and recognition process.Since it's time-consuming for ANN training, and there's much matrix-vector operation in it, it's fit for running the program on hardware with parallel computation capability, so bring forward a new method that implement RPROP algorithm by OpenCL C language, which runs on hardware with OpenCL support, to reduce the training time.In summary of Chapter 3 and 4, result of image processing and recognition to the website codes and its analysis are shown for further work.