Application Of Convolution Neural Network In Reproduced Laser Character Recognition

In the last few years,along with the proliferation of counterfeit and shoddy products as well as the improvement of counterfeit technology,the traditional anti-counterfeiting methods have been seriously affected.Now,the anti-counterfeiting center has developed a new kind of encrypted laser technology.The method is to press embossed patterns or characters on the aluminized layer.Only by radiating the identification with laser can we get the hidden patterns or characters.The reproduced anti-counterfeiting laser characters image means the hidden image we get from the laser irradiation.The research on character recognition is endless.However,few people work on the recognition of the specific form of the reproduced anti-counterfeiting laser text.In this paper,we will focus on the identification of such patterns and discuss how to improve the recognition rate.Taking into account the characteristics of the sample,such as low diffraction efficiency,the zero order interference to the first level image,the multi spot noise,the instability of the imaging and the limited number of samples,we preprocess the complicated pre-treatment of sample image,including grayscale,binarization and morphological processing,region segmentation and clipping.based on this,we adopts the method of convolution neural network to recognize the image,then analyzes the influence of the network depth,the size of the convolution kernel,the number of convolution kernels,and the type of the excitation function on the recognition rate,and we get an optimal model.In order to further verify the convolution neural network algorithm in laser reproduced of character recognition in the universality,this paper has conducted further experiments in the letter images,while using the traditional template matching method was compared.The experimental results show that convolution neural network has more advantages compared to the traditional template matching method.The traditional template matching method can not accurately extract the local features of the samples,and the recognition effect depends on the effect of pretreatment and the quality of template selection,compared to the traditional template matching method,it has a higher recognition rate and applicability.Base on it,we design and build a laser anti-counterfeiting characters recognition system in Matlab and Android platform.The system is simple,interactive and practical.