Research On The Algorithm For On-line Defect Detection Of Button Characters

The button is an important part of clothing. In the manufacturing process of button,there will be printed issue on the characters of button because the problem of printing crafts.At present, the defect of button characters is detected by human vision. But by human detecting,there will be the slow detection rate,low efficiency and high error detection rate,that is unable to meet the current requirements of modern management industries. So it is eager to use machine vision to replace human vision.According to this requirements,this thesis propose a suitable algorithm.Now,the characters recognition is mainly to solve the problem of text row-like arrangement.This thesis proposed a algorithm to solve the problem of text annular arrangement.By processing of buttons globally,we can get the position of the button through tracing boundary and finding inside holes.Then we can move the image by the ways of Log-polar transformation to get a image in which the characters are in a consistent order.By projection method and the algorithm of connection domain dividing,we can get c single character.Then we take the feature of button characters by grid method and strokes direction method.We use SVM to learn the feature to get classification. On the basis small number of sample,we use the classification of grid feature to test the integrity of the button characters.The correct rate is higher than 97%.We use the classification of strokes direction feature to recognize the button characters.The correct rate is higher than 91%.