Further Study On Workpiece Characters Recognition Algorithm For Casting DR Images

Non-destructive testing system can see the internal structure, its working principleis X-ray Digital Radiography (DR) for Casting is able to detect the internal defects ofcastings. It is particularly important to record the workpiece number of the defectivecastings workpiece for the subsequent quality assessment and product returns in thedetection process. When batch processing castings,Automatically recognize theworkpiece numbers through computer is able to overcome the affection of traditionalvisual recognition by the subjective and objective factors, thus it can speed up thedetection speed. The accuracy and recognition speed of workpieces characterrecognition is the key indicators that whether the technology can be applied to therecognition of the actual casting artifacts or not. So, It is very important to come up witha rapid and accurate recognition method. This article researches in thepreprocessing,skew correction,character segmentation and the character recognition.Ultimately, determine a series of algorithms to recognize the workpieces characterswhich are based on DR image.It need image preprocessing which includes enhancement and binarization beforerecognizing characters of casting. Workpieces. The quality of DR image-based castingworkpiece image is poor, the characters and background can not distinguish obviously,so we take the Gamma correction enhanced to highlight the workpiece character. Andthen use the binarization methods based on quadratic edge extraction, finally get a clearbinary image.In the DR testing process, in order to facilitate X-ray penetration railway castings,the castings will have a certain angle of rotation and swing,so the image of theworkpiece will also have a certain tilt. It will have a serious impact to latter part of theworkpiece character recognition. Therefore, this paper analyzes the common tiltcorrection methods for character recognition: such as the Hough transform method, themain direction of France and so on.Because of the casting workpiece number imagehave without a border information, and its noise is large and uneven.We will use theaxis of inertia and the Radon transform to do horizontal tilt and vertical tilt correctionfor the workpiece number. Then deburr and fill the corrected image. The result shows,this method can more accurately correct the image of the tilt of the workpiece numberCharacter segmentation contains vertical split and horizontal split.First,We propose a valid character vertical segmentation method to split a single character,according tothe width and the relatively fixed spacing of characters of the workpiece number.Then,according to the character of the horizontal projection on the character levelsegmentation. And normalize the single character sizeThis paper analyzes the used template matching, neural networks, support vectormachine (SVM) and Adaboost, the four character recognition methods, and studied anworkpiece character recognition method based on SVM. The support vector machinebased on statistical theory have better promotion and can effectively overcome the curseof dimensionality. It can be used to solve the practical problems of the small sample,nonlinear, high dimension and local minima points. At present, the SVM recognitionmethods mainly is applied in the license plate character recognition, handwrittencharacter recognition and so on. SVM is to solve two classification problems.However,the image railway of castings workpiece based on the DR include "0"-"9".For thisclassification, After the400-dimensional feature extraction, we use the one-on-oneclassification character recognition methods.Ultimately recognize the characters by theballot.The experimental results show that the method has a higher character recognitionrate and better robustness, and can basically meet the casting workpiece to identify theactual application requirements.