Research On The Application Of Deep Learning In The Anti-counterfeiting Of Dot Matrix Code Mixed Image Recognition

With the continuous increase of commodity types,counterfeit products emerge in an endless stream,resulting in serious harm.At the same time,a variety of anticounterfeiting measures are emerging to protect the safety of commodities.However,the traditional anti-counterfeiting methods have more or less shortcomings.In order to meet the anti-counterfeiting performance,production cost and convenience of use at the same time,this thesis selects the lattice anti-counterfeiting code as the anti-counterfeiting tool.In view of the shortcomings of lattice anti-counterfeiting code,this thesis proposes the encryption of data information and the improvement of unit coding module,and applies the improved anti-counterfeiting code to the trademark.At the same time,this thesis designs a kind of mixed code,through the recognition of dot matrix anti-counterfeiting code and super-resolution reconstructed image based on residual neural network,and introduces the deep learning into the application research of image anti-counterfeiting,completes the dual anti-counterfeiting of goods,and strengthens the anti-counterfeiting ability of goods.In this thesis,the purpose and significance of anti-counterfeiting technology research are described firstly,and the relevant literature about the current situation of deep learning in image super-resolution reconstruction and dot matrix code anti-counterfeiting is obtained by consulting the literature.After analyzing and summarizing the current situation,it points out the advantages and disadvantages of selecting lattice anticounterfeiting code.In order to solve the problem of coding security code in dot matrix security code,an asymmetric encryption method of RSA is adopted.The ciphertext is generated by encrypting the original content data,and the ciphertext data is transformed into the corresponding code graph.On this basis,a new coding module is proposed,which increases the amount of data information that the unit module can hold.Then,the lattice anti-counterfeiting code is transformed properly and integrated into the trademark information.In order to hide the dot matrix anti-counterfeiting code better in the trademark,the original positioning module is transformed into data module,and the trademark feature information is used to transform,and then the anti-counterfeiting code is identified.According to the needs of the company,a hybrid code is designed.This method combines dot matrix anti-counterfeiting code with image to generate mixed code.When the mixed code is coding,the original image is sampled down so that part of its details are lost and unrecognized.The unrecognized low-resolution image is mixed with the dot matrix anti-counterfeiting code to form the mixed code.When decoding,the dot matrix anti-counterfeiting code and the image after super-resolution reconstruction are identified successively,and the decoded information of the two is combined with the commodity information to determine the authenticity of the commodity.Experiments show that the improved dot matrix anti-counterfeiting code has the performance of information transmission and anti-counterfeiting,and can achieve anticounterfeiting at a lower resolution;through experiments,it is verified that the dot matrix anti-counterfeiting code which removes the positioning module can be hidden in the trademark,and can still be decoded through the trademark positioning;and the feasibility of the hybrid code is proved by experiments on the hybrid code.