利用捲積類神經網路定位複雜背景中的條碼

Abstract

條碼長期被當作資訊的圖形辨識原件，不過在複雜背景中，可自動偵測出不同扭曲或傾斜的條碼，還是一大挑戰。此研究提出可自動偵測這些類型的條碼定位系統。在這研究中用來測試此系統的條碼，包含一維條碼Code 39、Code 128和EAN- 13，與二維條碼QR code。此定位系統利用捲積類神經網路(Convolutional neural network)演算法，辨別影像中條碼的區域。接著透過影像處理的方法，將區域中的條碼切取出來。實驗結果證實此條碼定位系統是可以偵測特定範圍的條碼大小，甚至對於模糊或變形的條碼也能有效的偵測能力。此演算法在449張實驗影像中，可以達到86.25%的偵測率與78.55%切取率。

Barcodes have been long used for data storage. Locating barcodes in images of complex background is an essential yet challenging step for automatic barcode reading. This study aimed to detect and to extract one-dimensional Code 39, Code 128, and EAN-13 barcodes and two-dimensional QR barcodes in images of arbitrary backgrounds. The proposed method involved a convolutional neural network for detecting parts of barcodes. Once positive detection was confirmed, image processing algorithms were implemented to extract barcodes from the image. Experiments demonstrated that the proposed approach was able to locate barcodes of various module sizes and was robust to blurring, rotation, and deformation. The approach achieved an overall detection rate of 86.45% and an extraction rate of 78.55% using a set of 449 images.