降质先验引导的二维条码超分辨率研究  

Degradation prior guided quick response code super-resolution

作　　者：刘啟瑞 王晨 郭锋 林晨浩 岳焕景 沈盛 杨敬钰 LIU Qirui;WANG Chen;GUO Feng;LIN Chenhao;YUE Huanjing;SHEN Sheng;YANG Jingyu(Newland Auto-ID Tech,Fuzhou 350015,China;School of Electrical and Information Engineering,Tianjin University,Tianjin 300072,China)

机构地区：[1]福建新大陆自动识别技术有限公司,福建福州350015 [2]天津大学电气自动化与信息工程学院,天津300072

出　　处：《微电子学与计算机》2024年第7期18-28,共11页Microelectronics & Computer

基　　金：国家自然科学基金(62072331);福州市科技重大项目(2021-ZD-215)。

摘　　要：二维条码作为一种能够存储大量信息的图形标识,在工业生产的自动化控制、物流管理、质量追溯以及运输的信息交换等环节起着非常重要的作用。二维条码的高精度识别是实现快速、准确信息交换的基础。但是受拍摄环境和拍摄设备精度的制约,经常由于分辨率低而无法正确识别。针对此问题,提出了一种面向实际降质二维条码的超分辨率重建算法。考虑到实际降质的复杂性,提出了基于降质先验的超分辨率算法。首先,设计了一个降质先验信息编码器,用于提取和编码因拍摄环境和设备限制导致的图像质量降低的相关信息。然后,提出了一个降质先验引导模块,使用编码器提取的信息来引导主体结构的特征重建,包括降质特征图引导与降质先验引导两部分。由于目前缺少相关数据集,所以率先构建了真实退化条件下的二维条码超分辨率数据集(包含4944对低分辨率-高分辨率二维条码图像)。考虑到真实数据对之间有轻微位移,引入位移不敏感的损失函数对网络进行优化。实验表明,所提方法重建结果的峰值信噪比(Peak Signal to Noise Ratio,PSNR)、结构相似性(Structural Similarity,SSIM)以及扫出率这3种指标均优于5种经典的超分辨率重建算法,充分说明了所提方法的优越性。QR codes,as a graphic representation capable of storing a large amount of information,play a very important role in various industrial applications,including automation control,logistics management,quality tracing,and information exchange during transportation.High-precision recognition of QR codes is fundamental for achieving fast and accurate information exchange.However,due to the constraints imposed by the capture environment and the precision of capturing devices,low resolutions often hinder correct recognition.To address this issue,a super-resolution reconstruction algorithm for the real degraded QR code is proposed.Considering the complexity of real degradation,a degraded prior-based super-resolution algorithm is proposed.Firstly,a degraded prior information encoder is designed to extract and encode relevant information related to the degradation of image quality caused by capture environment and device limitations.Then,a degraded prior guidance module is proposed,which uses the information extracted from the encoder to guide the feature reconstruction of the main structure,including degraded feature map guidance and degraded prior guidance.Due to the lack of relevant datasets,a real degraded QR code super-resolution dataset is first constructed,consisting of 4944 pairs of low-resolution and high-resolution QR code images.Considering the slight displacement between real data pairs,a displacement-insensitive loss function is introduced to optimize the network.Experimental results demonstrate that the proposed method outperforms five classical super-resolution reconstruction algorithms in terms of Peak Signal to Noise Ratio(PSNR),Structural Similarity(SSIM),and recognition rate,indicating the superiority of the proposed method.

关 键 词：二维条码超分辨率 降质先验信息 降质先验引导模块 二维条码数据集 

分 类 号：TK448.21[动力工程及工程热物理—动力机械及工程]