Research On Spray Code Character Recognition Technique Of Plate

To overcome the plate spray character recognition issue under the extreme metallurgical environment, the dissertation proposes a machine vision based solution. Firstly, based on systematically character recognition technology, the framework of the recognition system and the code recognition process were designed under analysis the background and requirement of plate spray character recognition; Secondly, in view of the plate character image, image process method such as image pre-processing methodology、character regional location、character segmentation and feature extract were taken into account. Finally, The pattern recognition method combining the decision tree and SVM (support vector machine) were used for realizing the recognition result.This dissertation focuses on the following research directions:(1) character recognition process and the framework of the recognition system design focus on the demand of plate spray character recognition.(2) Use the image preprocessing method based on image graying, median filtering, background differenentiation, nonlinear histogram transform to get rid of the influences of the extreme metallurgical environment, iterative OTSU method were employed for image binarization and extract character region.(3) Analysis the characteristics of the whole character region, adopts the character region tilt correction method of least squares fitting and rotation projection. design of the character segmentation method, achieve accurate segmentation of the adhesion and fracture character.(4) an effective algorithm is selected for the normalization of single character image, and extract the statistical and structure features, compress feature based on PCA(5) For solving the recognition problem of multiple format character, research on multi-class classifier based on SVM, adopt the method based on decision tree sort out template style of the identify printing image. Use the candidate SVM classification based on multiple classifier to extract candidate of character. Take strengthens the detail of extraction structure feature to train one against one SVM classifier, get the recognition results of the candidate character, improved the recognition rate.