CAJ | 학술논문

为解决传统人工条烟复核方法与现有高速自动卷烟分拣系统不匹配等问题,基于形态学处理算法,利用条烟条码唯一性,提出了一种条烟图像自动识别方法.采用正方形自适应结构元素形态学算法处理二值化图像,减少条码内部条空区域对连通域的干扰,得到多个候选子连通域;采用子区域筛选方法定位条码区域,通过投影法定位可识别的字符并分割数字,完成条烟信息识别.以不同姿态不同品种的条烟图片在Matlab中进行仿真实验,测试条码定位分割算法的鲁棒性.结果表明:①该方法能够有效避免条烟表面字符、图形信息及光照噪声带来的干扰.②投影法可将供人识别字符与条空区域快速分割开,实现数字的有效定位,对于条码倾斜、低像素图像的数字分割效果良好.③加权模板匹配法结合模糊判别准则的数字识别算法,对于低品质数字以及易混淆数字均具有较好的识别效果.④在现场测试条件下,系统识别效率可达2 686次/h,准确率达95.2%.在满足系统要求下,采用该方法能够大幅提升条烟图像识别效率和准确性.
위해결전통인공조연복핵방법여현유고속자동권연분간계통불필배등문제,기우형태학처리산법,이용조연조마유일성,제출료일충조연도상자동식별방법.채용정방형자괄응결구원소형태학산법처리이치화도상,감소조마내부조공구역대련통역적간우,득도다개후선자련통역;채용자구역사선방법정위조마구역,통과투영법정위가식별적자부병분할수자,완성조연신식식별.이불동자태불동품충적조연도편재Matlab중진행방진실험,측시조마정위분할산법적로봉성.결과표명:①해방법능구유효피면조연표면자부、도형신식급광조조성대래적간우.②투영법가장공인식별자부여조공구역쾌속분할개,실현수자적유효정위,대우조마경사、저상소도상적수자분할효과량호.③가권모판필배법결합모호판별준칙적수자식별산법,대우저품질수자이급역혼효수자균구유교호적식별효과.④재현장측시조건하,계통식별효솔가체2 686차/h,준학솔체95.2%.재만족계통요구하,채용해방법능구대폭제승조연도상식별효솔화준학성.
To match a carton check system with the throughput of the high-speed automatic cigarette sorting system, an automatic carton image recognizing method based on morphological processing algorithm was proposed on the basis of unique carton bar code. The binarized images were processed by the square self-adaptive structure element morphological algorithm to minimize the interference of blank zones in a barcode on connected domain and obtain several candidate connected subdomains. By subrange filtering method to position barcode range, by projection method to position the discernible characters, segment figures and complete the recognition of carton information, the simulation experiment was conducted with pictures of cigarette cartons of different brands and randomly placed in Matlab to test the robustness of barcode positioning and segmentation algorithm. The results showed that: 1) The method effectively avoided the interferences brought about by character, image information and illuminating noises from carton surface. 2) Projection method segmented visible characters from blank zones in barcode quickly, implemented effective figure positioning and performed well in figure segmentation for oblique barcode, low pixel image. 3) Low quality or indistinguishable figures were well recognized by weighted matching algorithm combined with fuzzy criterion number recognition algorithm. 4) On-site testing showed that the recognition efficiency of the system reached 2 686 times per hour with the accuracy of 95.2%.