扭曲二维码的识别算法研究与实现

发布时间：2018-04-13 01:11

本文选题：二维码识别 + 复杂背景　；参考：《复旦大学》2013年硕士论文

【摘要】：二维码是用某种特定的几何图形按一定规律在平面分布组成黑白相间的图形以记录数据符号信息。因其有信息容量大、容错能力强、成本低等特点,被人们广泛应用电子商务、产品溯源等领域。在电子商务中二维码嵌入到广告背景,方便用户扫描二维码得到相应产品的详细信息或下载优惠卷；而产品溯源是在商品上张贴二维码,用户扫描了解商品生产流通信息。根据二维码在电子商务领域的应用,本文提出了一种基于形态学复杂背景下的二维码识别算法；根据二维码在产品溯源方面的应用,提出了扭曲情况下的二维码识别算法。无论是复杂背景还是扭曲情况,二维码识别包括二维码定位、二维码校正、译码三部分。复杂背景下的二维码由于背景干扰因素多,二维码定位是关键。二维码通过位置探测图形定位,传统的位置探测图形定位算法通过判断某些区域是否在水平和垂直方向满足1：1：3：1：1比率。为了排除复杂背景的干扰,我们在原有位置探测图形定位基础上通过位置探测图形的面积比和重心检测从众多候选区域中提炼出真正的位置探测图形,进而结合区域增长算法提取二维码区域。最后,通过角点检测及反透视变换得到校正二维码图形。扭曲情况下的二维码识别由于二维码发生非线性扭曲,二维码校正是关键。在通过前述二维码定位算法获得二维码区域后,本文提出了非线性扭曲二维码校正算法。此方法首先根据改进的直线逼近及位置探测图形区域增长获得扭曲二维码的角点及部分位置探测图形角点做为对应点,然后通过搭桥求出代表二维码边缘轮廓的关键点结合曲线拟合得到四条边缘的拟合曲线,再运用微积分知识计算出扭曲二维码上的对应点在校正图形上的坐标,最后通过非线性变换关系,得到校正图像。经过对100幅复杂背景下二维码图像试验,使用本文算法成功识别了95幅图像,相比传统算法识别率提高了10个百分点,在一定程度降低了复杂背景对于二维码识别率影响；对100幅扭曲二维码图像试验,成功识别了90幅图像,实验结果表明,本文自行设计的扭曲二维码识别算法,准确度高,具有现有软件及算法不具有的非线性扭曲图像识别功能。
[Abstract]:Two-Dimensional Code (QR) is used to record the data symbol information by using a certain geometry to form a black-and-white pattern in the plane according to a certain law.Because of its large information capacity, strong fault tolerance, low cost and so on, it has been widely used in electronic commerce, product traceability and other fields.In electronic commerce, the QR code is embedded in the advertising background, which is convenient for users to scan the QR code to get the detailed information of the corresponding product or to download the preferential volume; the product traceability is to post the QR code on the commodity, and the user can scan and understand the information of the commodity production and circulation.According to the application of QR codes in the field of electronic commerce, this paper presents an algorithm for the recognition of QR codes based on complex morphological background, and an algorithm for recognizing QR codes under distorted conditions according to the applications of QR codes in the traceability of products.Whether in complex background or distorted case, QR code recognition includes two dimensional code location, two dimensional code correction and decoding.Due to many background interference factors, two dimensional code location is the key in complex background.The traditional location algorithm determines whether some regions meet the ratio of 1: 1: 1: 1: 1 in horizontal and vertical directions.In order to eliminate the interference from complex background, we extract the real position detection pattern from many candidate regions by the area ratio and gravity detection of the position detection pattern based on the original position detection pattern location.Furthermore, the region of two dimensional code is extracted based on the region growth algorithm.Finally, the corrected QR code figure is obtained by corner detection and inverse perspective transformation.Due to nonlinear distortion of two dimensional code, two dimensional code correction is the key.After obtaining the two dimensional code region by the above two dimensional code location algorithm, a nonlinear distorted two dimensional code correction algorithm is proposed in this paper.Based on the improved linear approximation and the growth of the position detection graphics region, the corner points of the distorted two-dimensional code and the corner points of the partial position detection graphics are obtained as the corresponding points.Then the key points representing the edge contours of the two-dimensional code are obtained by bridging and the fitting curves of the four edges are obtained by fitting the curves. Then the coordinates of the corresponding points on the distorted two-dimensional code are calculated by using the calculus knowledge.Finally, the corrected image is obtained by nonlinear transformation.After the experiment of 100 images with complex background, 95 images are recognized successfully by using this algorithm. Compared with the traditional algorithm, the recognition rate is increased by 10 percentage points, and the influence of complex background on the recognition rate of two-dimensional code is reduced to a certain extent.90 images have been successfully identified from 100 distorted two-dimensional code images. The experimental results show that the algorithm designed in this paper has high accuracy and has a nonlinear distorted image recognition function which is not available in existing software and algorithms.
【学位授予单位】：复旦大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP391.44

【相似文献】