以資訊植入及深度學習提升圖像化二維條碼實體輸出的辨識能力之研究
No Thumbnail Available
Date
2021
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
QR code 是目前最普遍被採用的二維條碼，由於其為黑白模塊所組成，影響視 覺美觀，且在列印輸出時，因尺寸大小、網點擴張等印刷條件因素，導致條碼資訊 容易失真，影響解碼辨識。為了能夠將印刷輸出之小尺寸美化 QR code 保持視覺美 觀並且穩定解碼，因此本文提出了一套系統性的圖像化 QR code 資訊植入技術，列 印後掃描將辨識結果進行錯誤分析，了解 QR code 之黑點與白點資訊點模組的錯誤 特性並加以改善，最後以深度學習辨識來進行錯誤分析。實驗結果顯示，本研究所 發展的方法能相容於現行的列印輸出設備，在調整白色資訊點的植入訊息強度後， 可有效抑制因網點擴張所造成的「偽黑」 辨識錯誤的情形。且輸出的小尺寸圖像化 QR 仍有較佳視覺品質，降低錯誤發生率，並藉由深度學習辨識提升辨識能力，有 效增進美化 QR 的成功讀取率。對於彩色影像在指定輸出裝置的條件下，可得到最 佳化的 QR code 植入訊息方法及讀取能力，未來能夠運用於商業加值應用上，並彰 顯實體輸出條件對於圖像化 QR code 整合應用的重要性。
QR codes are currently the most commonly used 2D barcodes, composed of black and white modules, which detracts from their aesthetic appeal. When printed, due to size, dot gain, and other printing conditions, barcode information is easily distorted, yielding poor recognition results. In order to beautify the small size of the printed output QR code to maintain visual beauty and stable decoding. We present a systematic aesthetic QR code information embedding technique as well as an error analysis method for physically printed QR codes. Understand and improve the error characteristics of the black dots information and white dots information of the QR code, and finally perform error analysis with deep learning recognition. The experimental results show that the proposed method is compatible with current printing and output equipment. Judicious adjustment of the embedded strength of white module dots decreases the false-black recognition error caused by dot gain, and yields small printed aesthetic QR codes that look better. This improves the decoding rates of aesthetic QR codes, and through deep learning recognition to improve recognition ability. We optimize the QR code embedded method and reading ability given specified output device conditions. This highlights the importance of output conditions for integrated applications of aesthetic QR codes.
QR codes are currently the most commonly used 2D barcodes, composed of black and white modules, which detracts from their aesthetic appeal. When printed, due to size, dot gain, and other printing conditions, barcode information is easily distorted, yielding poor recognition results. In order to beautify the small size of the printed output QR code to maintain visual beauty and stable decoding. We present a systematic aesthetic QR code information embedding technique as well as an error analysis method for physically printed QR codes. Understand and improve the error characteristics of the black dots information and white dots information of the QR code, and finally perform error analysis with deep learning recognition. The experimental results show that the proposed method is compatible with current printing and output equipment. Judicious adjustment of the embedded strength of white module dots decreases the false-black recognition error caused by dot gain, and yields small printed aesthetic QR codes that look better. This improves the decoding rates of aesthetic QR codes, and through deep learning recognition to improve recognition ability. We optimize the QR code embedded method and reading ability given specified output device conditions. This highlights the importance of output conditions for integrated applications of aesthetic QR codes.
Description
Keywords
QR code, 圖像化二維條碼, 印刷, 資訊植入, 深度學習, 卷積神經網路, QR code, Aesthetic QR code, Output, Information Hiding, Deep Learning, Convolution Neural Network