Verified Code Recognition Algorithm Based On SVM

In the early 1990s, the support vector machine (SVM) was first propounded by Boser, Guyon and Vapnik, which is a new generation of learning systems based on recent advances in statistical learning theory. The SVM has high generalization performance of learninge in real-world applications such as text categorisation, hand-written character recognition, image classification, biosequences analysis, etc. With the continuous deepening in the theory and the continuous development in practice, the SVM are now established as one of the standard tools for machine learning and data mining.Verified code recognition based on SVM is widely used in e-commerce. The development of e-commerce needs a certain way to promote its goods and disabilities need urgent attention; thus, many people began to study the web robot technology. They hope that it can realize automatic registrations for email systems, multi-send messages, automatic flood-blogging and auto-login, et al. These functions are inseparable from the verification code recognition technology. So the verified code recognition technology has become a hot research topic in this field.In this thesis, we introduce the basic theory on SVM, algorithm implementation strategies, models and parameter selection, studie the validation code image processing and feature extraction methods, and illustrate the specific steps of the application of SVM to the code identifying. We design a set of practical verification code identification program and obtain good results in the code recognition based on SVM.First of all, the statistical learning theory and the SVM construction methods are introduced in this thesis, and the multi-class SVM classification algorithm and the kernel-based methods are studied and analyzed.Secondly, the various basic image processing algorithms are introduced, some validation code image processing problems are analyzed and solved to achieve better results and facilitate code feature vector extraction. Moreover, several different verification code feature extraction methods are analyzed and compared.Thirdly,the detailed design process of the multi-class SVM classifier is introduced. The implementation process conditions are analysed, a data structure is designed to store the vector data, and an interface of training and prediction is provided to train the SVM and predict the results of classification, respectively.Finally, based on the above theory and using C++,the SVM algorithm is presented to realize a verification code identification system. Tests on the typical verification codes downloaded from a variety of different web sites show that the proposed system can efficiently identify all types of popular verification codes in network.