【摘要】 目的针对包装产品外壳上黑白QR码易受到污渍侵蚀损坏,长期磨损易模糊,以及图像采集过程易出现失焦模糊、运动模糊,导致无法完成识别需求,提出一种基于栈式降噪自编码器的受损QR码恢复的预处理方法,达到显著修复包装产品上受损的QR码图像并提高其识别率的目的。方法通过深度学习模型栈式降噪自编码器,可以将受到噪声干扰的像素点根据受损像素数据映射到以标准数据为参照的高概率数值点,实现整个受损QR码基于像素点的重构恢复,从而提高识别率。结果通过对实验QR码进行高斯模糊、随机污渍侵蚀等多种方式的损坏,能够将识别率较低或完全不能识别的测试图像集恢复出高质量的QR码图像,显著地提高了识别率,并且速度快、可重复性好。结论采用基于栈式降噪自编码器的受损QR码恢复的预处理方法,能够重建受损的QR码,并可以广泛应用于包装产品QR码识别前的预处理,以提高识别率。更多还原

【Abstract】 The work aims to propose a pretreatment method based on stacked de-noising autoencoder for the recovery of damaged QR codes, to significantly repair the damaged QR code image on the packaging products and enhance the recognition rate, with respect to the problem that the black-and-white QR code on the shell of the packaging product is prone to be stained, eroded and damaged, and become fuzzy due to long-term wear, and easily subject to out-of-focus blur and motion blur in the process of image acquisition process, leading to the inability to complete identification requirements. Through deep leaning of stacked de-noising autoencoders of the model, the pixels interfered by the noise could be mapped to the high probability numerical point with standard data as reference according to the damaged pixel data, so as to achieve the reconstruction and restoration of the whole damaged QR code based on the pixels, thus improving the recognition rate. Through damages to the experimental QR codes, including Gaussian blur, random stain and erosion, etc., the test image set with low recognition rate or completely unable to be identified could be restored as the high-quality QR code image, significantly improving the recognition rate, featured by fast speed and good repeatability. The pretreatment method based on stacked de-noising autoencoders for the recovery of damaged QR codes can restore the damaged QR code, and be widely applied in the pretreatment before the identification of packaging products’ QR code, in order to improve the recognition rate.更多还原