Embossed Characters Detection And Recognition Based On Vision

Embossed characters are widely used in production and manufacturing,and are very important tools in information carrying and transmission.The embossed characters are made by mold casting,engraving or other methods to produce uneven deformation on the surface of the object.The color and material of embossed characters are similar with their background,and they are three-dimensional characters.Due to this remarkable feature of the embossed character,its recognition becomes difficult.With the improvement of modern industrial manufacturing and production efficiency,the automatic recognition of embossed characters has become an inevitable requirement for the intelligent production of industrial products.The characteristics of the embossed character cause it to be different from the ordinary optical character recognition(OCR)method,and the collection conditions are high,so it is impossible to use the existing mature OCR method.In this paper,the following problems are that it is difficult to recognize embossed characters.According to the image of EVA material sole,the corresponding acquisition hardware system is designed.First,the composition of the hardware acquisition system is briefly described,the performance parameters and selection methods of the relevant hardware composition are analyzed.Then the appropriate hardware is selected according to the research object of this paper and the requirements of the vision acquisition system.The characteristics of the embossed characters of the sole and the characteristics of the light source are analyzed.This paper discusses the influence of illumination on the embossed characters.So the optimal lighting scheme of the recognition system is designed as follows: the strip white light source is placed on the side of the light,which lays the foundation for the subsequent algorithm with good foundation.The enhancement method for embossed characters is studied: an embossed character enhancement algorithm based on convolutional neural network(CNN).First,the commonly used image enhancement methods are introduced,and then the defects and deficiencies of the traditional enhancement methods in processing the characters are analyzed according to the characteristics of the embossed characters.Then combined with the relevant knowledge of deep learning,a new embossed character enhancement method based on CNN is proposed for the embossed character of the sole.The method of preprocessing embossed characters for soles is studied which is a special carrier,combining the characteristics of soles,it was necessary to remove round holes.After enhancing the characters,because there are many holes in sole,which makes it difficult to recognize.It is proposed to remove the holes based on the combination of CNN and image pixels.Experiment result achieves the ideal pretreatment effect of the embossed character of the sole,and lay the foundation for subsequent recognition.The embossed character recognition algorithm based on Tesseract is studied.After enhanced and further pre-processed,the embossed characters can be recognized by the classic OCR algorithm.Based on Tesseract's character recognition method,the embossed character samples are recognized,and the characters can be detected and recognized more accurately.This paper combines the machine vision-based embossed character preprocessing method and OCR algorithm to solve the problem of long recognition time and unstable accuracy of the traditional embossed character,which also achieves a better recognition rate.