Efficient 1D Barcode Localization Method for Imagery Shipping Label Using Deep Learning Models
Lang-chao Qiao, Jin-lu Wang, Bao-hong Gao, Xingang Yang, Wentao Feng, Yuxiao Zhang, Yan Wang, Hai Liu, Wei Liu · 2021
In this paper, an efficient approach for imagery shipping label 1D barcode localization integrated with two deep learning models is proposed. In the first stage, a Faster R-CNN model is adopted to detect the barcode region existed in the imagery shipping label automatically. Then orientation of parallel lines of 1D barcode can be computed by using Line Segment Detector algorithm. Next, the tilt angle of the image will be corrected using the direction information for rotation correction. Finally, in order to decode barcode accurately in the subsequent stage after barcode localization, a Resnet-34 model was used as a multi-class classifier to determinate the orientation of the shipping label and calibrate the whole image. Using a dataset that contains a total of 1830 imagery shipping labels including slanted ones or ones with complex background, our method can yield a barcode localization accuracy of 0.9866, which demonstrates its encouraging performance.