Research Of Laser Anti-counterfeiting Code Recognition With Complex Background

The phenomena of shoddy, fake and changing in sales of tobacco in China havebrought great huge economic and make bad influence in sales. Anti-counterfeitingcode of cigarettes which provides a lot of important information such as delivery date,sales area, and customer number can prevent the fake, shoddy and sales in wrong area.For these reasons, research on the recognition of laser anti-counterfeiting code withcomplex background is needful and significant.While traditional methods for Optical Character Recognition (OCR) are mainlyapplicable for high-quality and low-noise text images, the performance of methodswill get decreased when the image quality declines. Several techniques forrecognition of laser anti-counterfeiting code with complex background are proposedin this paper. Firstly, a threshold method based on Ensemble Empirical ModeDecomposition (EEMD) is applied to binarize images in the image pre-process stage.Secondly a robust method for features extracted utilizing the Log-Gabor filters isproposed. Log-Gabor transform will extract the character image’s local texturefeatures without the illumination, noise and stroke distortions impacts. This paperconsists of3sections:(1) Discussing the characteristics of anti-counterfeiting images. And enhancingthe character features via filtering noises in the image. After the image histogramanalyzed by EEMD, we computed the value of threshold depend on the result ofconstruction with Intrinsic Mode Function (IMFs) and image characteristics.According to these results of experiment, it can be concluded that the proposedmethod is really competent to bnarize the images efficiently.(2) Designing Log-Gabor filters to extract character features. The theory ofGabor and Log-Gabor transform is presented firstly. And then we examined thefeature-extracting method based on Log-Gabor filters for anti-counterfeiting coderecognition in binary images and gray-scale images. The results show the accuraciesof different features extracted from the output of Gabor filters and Log-Gabor filters.The experimental results support that the method proposed in the paper has excellentperformance on anti-counterfeiting code recognition. (3) The theory of SVM classifier is presented, and a classifier foranti-counterfeiting code recognition is designed. In addition, we discuss theoptimization methods for parameters selection of classifier. And the optimized SVMclassifier under examination gets a higher accuracy.Recognition rate of98.05%and98.79%are achieved with our method for laseranti-counterfeiting code in binary and gray-scale images. The experimental resultsshow that our method performs effectively for low quality images.