Development Of Cosmetic Paper Label Defect Detection System Based On Machine Vision

Cosmetics,as a beauty product favored by women in today's society,its product information is especially important for consumers.Cosmetic label as a product identification method is an important indicator to measure product quality.The label describes the information of cosmetics composition,date of production,color,brand and other information,which are concerned by consumers,so the quality of cosmetic label is getting more and more attention.However,in the production process,due to factors such as production process and production environment,the labels produced often have many quality problems,such as the leakage,offset,tilt and overlay of adhesive labels,the leakage,offset,tilt,multiple spraying and date printing errors of the production date characters,so the label defect detection process is crucial.With the development of computer technology and digital image processing technology,automation technology based on machine vision is gradually replacing manual work.At the same time,in view of the current situation of manual detection of cosmetic paper labels and the actual needs of cosmetic label manufacturers,we developed a set of automatic detection system for cosmetic paper label defects,and proposed related technologies to solve the key problems of this topic.The specific contents are as follows:Image preprocessing algorithm design: mainly includes image noise reduction processing,tilt correction and label ontology image extraction.Through analysis and research,this paper proposes an improved label filtering enhancement method combining mean filtering and impact filtering.The hough transform and affine transformation are used to correct the tilted label ontology image.Finally,the label ontology image segmentation extraction is introduced.At the same time of introducing different segmentation methods,adaptive OTSU threshold segmentation is selected as the extraction method of tag ontology image.Label image position defect detection algorithm design : Firstly,for the position detection of the self-adhesive label,the surface information of the label image is analyzed,and the position of the sticker label is fixed relative to the image of the label body,and the gray value of the sticker label is lower than other surface details of the label image,and the maximum connected domain can be obtained after the threshold is divided.With this feature,this paper proposes an image moment-based self-adhesive label position defect detection algorithm;Secondly,for the position defect detection of date character printing,the analysis shows that the label date has a fixedcharacter "EXP" and the date character print position is uncertain.This paper proposes a date character position detection algorithm based on contour feature matching;Finally,after a large number of experiments,the results show that the position defect detection algorithm proposed in this paper can detect the position defects such as leakage,superposition,tilt,offset and date character leakage,tilt and offset of the sticker.The detection rate was 94.2%.Date character recognition algorithm design: In order to ensure the accuracy of the label production date and prevent the waste of large quantities of raw materials,this paper proposes a multi-character feature combination method to train the neural network to complete the identification of date characters.Firstly,the morphological closing operation is performed on the tilt-corrected characters,so that the free points in the lattice characters are connected into a line;Secondly,the date characters are segmented by the combination of vertical projection and horizontal projection;Thirdly,the neural network is trained by extracting mesh features,projection features and character duty cycle features from the segmented normalized characters.Finally,the date character is recognized by BP neural network and the recognized character is checked and compared.After a large number of experiments and literature comparisons,BP neural network trained by this method can efficiently identify date dot matrix characters with a recognition rate of 95.1%.Defect detection system software design: This defect detection system software uses the Sapera LT camera software library and the Halcon12.0 image processing library.Based on the Visual Studio 2010 development platform,the entire software system framework is built using MFC(Microsoft Foundation Classes).The design method developed a reliable and stable cosmetic paper label defect detection system.