Hcc-svm-based Character Recognition Technology

Developed on the basis of the statistical learning theory, Support Vector Machine (SVM) is not only a good machine learning algorithm based on data, but also a powerful data classification technology. It possesses strict theory and mathematic foundation. It has a wide range of applications in text categorization, character recognition, image recognition, face recognition, fingerprint recognition, protein structure, genetic test, etc.Focusing on character recognition technology, this paper has studied up on pre-processing algorithm for character recognition and support vector machine classification algorithm in depth. The research of the character image pre-processing focuses on the gray-scale image filtering algorithm, the skew detection of image and character segmentation algorithm; HCC-SVM has been presented as a kind of new fast SVM classification algorithm based on Hierarchical Clustering by Category (HCC).The main study contents and innovations of this paper are as follows:(1) From the view of pulsed noise-point detection, a new filtering method of gray-scale image has been proposed. When the noise density is equivalently high , it can also achieve better filter performance and faster speed.(2) Taking advantage of features that the first line of text is in line and has not interfere above, a skew detection method of text images has been presented. It could meet the needs of skew detection of plain text images in connection with a faster processing speed.(3) Taking advantage of the feature of text line interval, a character segmentation method of rows and columns has been adopted. It can achieve the character segmentation of text image when the quality of images is better.(4) Considering the matter of time and space produced when the traditional support vector machine processes large training data, HCC-SVM has been presented. Compared with traditional SVM algorithm, it can greatly improve the efficiency of algorithm. And compared with SVM algorithm based on random sampling, it can also achieve better performance of segmentation.(5) The key parameters of SVM algorithmhas been studied by used of the method of violence detection , and the parameters selection has been proposed to enhanced to satisfy needs of testing data sets.