Arabic Manuscript Author Verification Using Deep Convolutional Networks
2017 1ST INTERNATIONAL WORKSHOP ON ARABIC SCRIPT ANALYSIS AND RECOGNITION (ASAR)(2017)
摘要
In this paper, we propose an automatic method for manuscript author verification based on an analysis of consecutive patches extracted from an image. The classification algorithm uses a deep convolutional network with two types of patch extraction: one based on connected components and the other based on usage of a fixed-size sliding window. We apply this method to verify the authorship of the Arabic manuscript entitled al-Khitat attributed to the hand of the renowned medieval Arab historian al-Maqrizi. Using appropriately collected ground-truth labeled data for convolutional network training purpose, our method has demonstrated promising results when applied to previously unseen manuscripts.
更多查看译文
关键词
appropriately collected ground-truth labeled data,fixed-size sliding window,patch extraction,classification algorithm,deep convolutional network,Arabic manuscript author verification
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要