基于中心化Jaccard匹配的SD卡光学字符验证
2023,31(7):290-297
摘要：为了解决安全数码卡(SD卡,secure digital card)表面字符人工验证效率差、准确率低的问题,提出了一种基于中心化Jaccard匹配的SD卡光学字符验证方法,能够实现SD卡表面字符的精密检测与自动验证。首先,提出了一种基于HSV三通道直方图分析的快速验证方法,实现特征显著SD卡图像的准确验证；其次,针对SD卡字符验证精度受光照变化和微小旋转影响的问题,提取V通道图像和变化角度特征,提高HOG特征对光照和旋转变化的抵抗能力；最后,针对相似SD卡字符的验证问题,提出了一种中心化广义Jaccard系数,增强了相似度指标的辨别力,实现了特征相似图像的精密检测。以实际场景采集的数据对所提方法进行验证分析,试验结果表明,该算法准确率达到99.15%,具有很好的实用性和鲁棒性。
关键词：机器视觉;SD卡;光学字符验证;HSV三通道直方图;方向梯度直方图(HOG);广义Jaccard系数
SD Card Optical Character Verification Based on Centralized Jaccard Matching
Abstract：In order to solve the problem of poor efficiency and low accuracy of manual verification of surface characters on SD card(secure digital card), an optical character verification method based on centralized Jaccard matching was proposed, which can realize the precise detection and automatic verification of surface characters on SD card. Firstly, a rapid verification method based on HSV three-channel histogram was proposed, which can accurately detect SD card images with obvious apparent features. Secondly, aiming at the problem that the character verification accuracy of SD card was affected by illumination changes and micro rotations, the resistance of HOG features to illumination and rotation changes was optimized by extracting V-channel images and changing angle features, and the detection accuracy was improved. Finally, aiming at the verification problem of similar characters on SD card, a centralized generalized Jaccard coefficient was proposed, which enhances the discrimination of similarity index and realizes the precise detection of similar images. The proposed method is verified and analyzed based on the data collected in the actual scenario, and the test results show that the accuracy of the detection algorithm reaches 99.15%, which has good practicability and robustness.
Key words：machine vision; color inject character; optical character verification; HSV color space; HOG(histogram of oriented gradient); generalized jaccard coefficient
收稿日期：2023-02-10
基金项目：国家自然科学基金(62072416)；国家自然科学基金(61873246)。
