Cigarette packet seal defect detection based on image recognition technology of support vector data description

Abstract

Based on image recognition technology, a method of defect detection for cigarette packet seal sticker is proposed in this paper. Based on the proposed method, a defect detection system for cigarette packet seal has been developed. The captured image with CCD can be filtered and its edge can also be extracted with the developed system. The Radon transform was applied to detect the position and angle of the straight line of the extracted edge. Support vector data description with Gaussian kernel was applied to classify the defective packet seal stickers based on the feature of the captured image. The cigarette packet with seal defection can be detected and pushed out of the production line with the developed system. The long term application results in one factory of China Tobacco have verified the effectiveness and precision of the proposed method and the developed system.