以人工智能辨識高明度紙張紋理特徵

Using Artificial Intelligence to Identify Paper Texture Features

這張是什麼紙張？此為印刷業者常被問到的問題。單憑客戶拿著一張紙要找出是什麼紙張、哪些廠商有進口或生產這種紙張更是一大難題。如果能有好的智能辨斷方法，將來是有機會建立完善的紙張資料庫做為比對的基礎。因此本研究採用卷積神經網路（CNN）及局部二值模式（LBP）二種方法提取18種白紙特徵加以分析判斷是那一種紙張。其結果顯示二種方法均能達到99%以上的準確率，而以局部二值化的特徵加上隨機森林分類方法所花的時間最短。

What kind of paper is this? This is one of the questions often asked by printers. Therefore, this study uses two methods of convolutional neural network (CNN) and local binary pattern (LBP) to extract 18 kinds of white paper features to analyze and judge which kind of paper it is. The results show that the two methods can achieve more than 99% accuracy rate, and the local binarization feature with random forest classification takes the shortest time.