Research And Implementation Of Positioning And Registering Control System For Curved Screen Printing Image

With the deepening of global industrialization,the traditional way of relying solely on manual control of the color quality of curved screen printing has long been unable to meet the needs of automated production in the market.It is urgent for curved screen printing machine manufacturers to upgrade and transform their machines in order to achieve people’s higher demands for practical functions,aesthetics and grades of products.Registration accuracy is an important index to evaluate the quality of printing.In multicolor printing,because the overprinting method is directly printed on the curved substrate,it cannot design the registration mark,and is susceptible to printing machinery,environment and human factor.Factors such as operation cannot guarantee that the color and tone of the printed product reproduce the original well.Therefore,the on-line product quality inspection method based on machine vision technology is adopted to achieve high-quality multi-color registration effect,improve production efficiency and screen printing quality,give full play to the advantages of open market and low production cost of curved screen printing,and enhance enterprise competitiveness and product profit.The research contents of this thesis are as follows :(1)Investigate the development status of curved screen printing production at home and abroad,expound the basic theoretical knowledge of curved screen overprint process,and learn machine vision and digital image processing technology.In-depth analysis of the actual needs of printing enterprises and the market,the design of curved screen printing image positioning registration control system,divided into four layers of software architecture,select the adaptation of hardware equipment,the use of serial communication to transmit data,the multi-color overprint quality manual detection into machine vision control detection.(2)A multi-layer image localization algorithm based on phase correlation and feature point registration is proposed.The image is preprocessed by grayscale,median filtering,and open operation to eliminate redundant information.Based on the phase correlation algorithm in the frequency domain,the Hanning window is added to the image.The Fourier shift theorem is used to determine the relative translation of the two images,and the two images are adjusted to the same position according to the translation value.Image registration combination algorithm based on feature points,using OFAST algorithm to extract feature points and BRISK feature description algorithm to calculate descriptors,to obtain representative feature points;then,the Hamming distance is used to measure the similarity between binary descriptors,and the vector cosine similarity and PROSAC algorithm are combined to eliminate the mismatch,and then the ordinate displacement deviation of the optimized matching point pair is calculated.(3)Design user interface based on MFC to realize the visualization of operation,image and data.The average value of the displacement difference of the image registration method based on the feature points is obtained,and the geometric operation is performed with the relative translation value obtained by the phase correlation method.The final symbolic displacement difference is calculated and fed back to the lower computer to adjust the overprint deviation.(4)Set three objective indicators: accuracy,motor error and algorithm execution time to measure the performance of the algorithm and evaluate the registration ability of the algorithm.The experimental results show that the algorithm has small calculation amount,high calculation efficiency,stable registration effect,and good robustness to images with local deformation,missing printing,blurring,scattering,shadow and other disturbances.Qualitative observation has not found the wrong matching point pairs of the optimized registration effect images.The accuracy of quantitative analysis and calculation of translation value is more than 98%,the registration time of the whole image is less than 300 ms,and the step error is less than 1 pixel,which can meet the practical application requirements in complex scenes.