一种高精度的体育商标分割算法  

A HIGH PRECISION SEGMENTATION ALGORITHM OF SPORTS TRADEMARK

作　　者：岳英杰 刘盼[1] 徐荣青[1] Yue Yingjie;Liu Pan;Xu Rongqing(College of Electronic and Optical Engineering,Nanjing University of Posts and Telecommunications,Nanjing 210046,Jiangsu,China)

机构地区：[1]南京邮电大学电子与光学工程学院,江苏南京210046

出　　处：《计算机应用与软件》2022年第12期246-251,共6页Computer Applications and Software

摘　　要：在体育比赛中,俱乐部和赞助商希望观众实时获取宣传信息,可通过对运动员衣服上的商标识别来实现。图像分割是商标识别的一个关键部分,目前的分割算法对于背景复杂的图片分割精度不高,针对这一问题,提出一种高精度的商标分割算法。该算法融合了Graph Cut和Grab Cut两种算法的思想,利用Graph Cut将图片分为不同区域,通过设定双阈值判定不同区域属于背景或前景,在此基础上初始化图片的掩膜,再进行Grab Cut迭代,完成图像分割。实验结果表明,该算法与Grab Cut算法相比,有效解决了欠分割以及过分割的问题,并且保留了图像细节,克服了背景噪声,算法分割精度高达94%。In sports competition, the spectators must rely on the trademark recognition technology to obtain the club and sponsor information in real time through the trademark on the athletes’ clothes. Image segmentation is a key part of trademark recognition, but the accuracy of traditional image segmentation algorithm is not high. To solve this problem, an efficient trademark segmentation algorithm is proposed. This algorithm combined the ideas of graph cut and grab cut. The image was divided into different regions by graph cut, and the probability that the region belongs to the background and foreground was calculated by using double thresholds, so as to advance ahead. The pedestrian intervened to initialize the mask of grab cut, and performed the iteration of grab cut to complete the image segmentation. The experimental results show that compared with the grab cut algorithm, the algorithm effectively solves the problem of under segmentation and over segmentation, and effectively retains the details of the image, overcomes the background noise, and the segmentation accuracy of the algorithm is up to 97%.

关 键 词：商标 Grab Cut 图像分割 掩膜 双阈值 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]