Contour Identification for Label Cutting Machine Based on Sobel Edge Detector and Contour Tracing Scheme

• Main Authors: Hsu,Ting-Wei, 許庭瑋
• Other Authors: Yu,Chu
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/jbpc6a

碩士 === 國立宜蘭大學 === 電子工程學系碩士班 === 106 === In recent years, product customization is in great demand, such as custom labels. Because label customizing processes have the high computation complexity and long computation time, they are high cost to generate the label contour. This results in the cost reduction to be an important part for label customization. Unfortunately up to now, most of the contour images are generated by manual work, because the application program of identifying label contour is not yet available. Thus, the high-speed automatic label contour generation is required to significantly reduce the processing time of custom labels. In this Thesis, we develop an automatic contour generation technique for custom labels. Because the label image has lower noise and simpler structure compared with other generic images, we can use lower complexity algorithm to develop an efficient contour tracking method for the label image. Based on this reason, we adopt Sobel edge detector with lower computing time and well efficiency to reduce the amount of the data and to capture the feature for label images. After edge detection, the proposed method employs mathematical morphology to dilate the mistaken and miss edge. Moreover, this dilation process also agglomerates the dispersion object in multiple contour objects. After the aforementioned preprocessing, the proposed method traces boundary contour in a binary image to generate a closed-loop contour for the label image. Finally, compared with standard contour images depicted by technician and contour image generated by active contour model, the proposed method has better performance. In qualitative analysis, the new contour image generated by the proposed method is quite similar to standard contour ones. In quantitative analysis, the new contour image generated by the proposed method achieves at least 90% structural similarity index and has less computation time compared to standard contour images.