Identification Technology A Natural Scene Sign Characters

There exits lots of traffic signs in natural scenes,and the information of signs brings a lot of convenience to people’s daily life.An automatic tool developed to recognize road sign text from road sign images is of great value to visual navigation and scene understanding.This dissertation studies the related techniques for Road sign text extraction and recognition from road sign images.The main contents of dissertation are as follows.Firstly,this paper proposes a new method named "rectangle+Harris corner points" to locate road sign region from the image. For this method,first it detects all the rectangular outline in the image,then uses Harris corner detection algorithm to extract corners,final the rectangular contour area with a maximum number of Harris corner points is determined as road sign rectangular area.Verified by experiments,in general, the method for road sign rectangular area positioning has high accuracy.Secondly,in this dissertation,the preprocessing of road sign region image can be divided into two parts,including image distortion correction and character segmentation.For the part of image distortion correction,frist it finds four groups corresponding points in the image plane and the word plane,uses "four-point method" to correct the distortion of the image,and studies the related Interpolation algorithm which should be applied in image distortion correction. For the part of character segmentation,first it uses binarization algorithm to realize the segmentation of text pixels and background pixels in the image,then uses connected component labeling method and the further processing of the labeled image to achieve a single character image.The experimental results show that the method used by this dissertation can get very good segmentation results for cross-type, column type of road signs.Finally,for low-quality Chinese character,the method of two classifiers is proposed in this paper,first coarse classification and fine classification.In the process of classification,KNN classifier, always with high accuracy, is used.In coarse classificatio,it uses the link feature of coarse grid characteristics and the crude peripheral feature to built eigenvectors.In fine classification,the first step it constructs the block structure of the Chinese character image,total divided into five,and uses principal component analysis to extract characteristics for each sub-block image;the second step it builds five weak KNN classifiers based on five sub-block images;the third step combined with the results of five classifiers, it uses voting scoring method to determine the final recognition result.And we analyses the characteristics of a single Chinese character image which obtained by the segmentation.then use artificial methods to construct samples which rich in low-quality information, and thus establish a database.The experimental results have demonstrated that the method of two classifiers in this paper obtains a higher recognition rate for the low-quality of Chinese character recognition.