一种基于深度学习的图书封面文字自动检测识别系统

A Deep Learning Based System for Automatic Detection and Recognition of Book Cover Text

图书封面文字的检测和识别有助于快速获得图书的相关信息,但该问题在现实中容易受到复杂背景、光照、几何变形以及文字的不同字体、大小、粗细等干扰因素的影响。为此,文章提出一种三步算法。首先,通过Hough变换对包含图书封面的图像进行旋转矫正;然后,通过CRAFT神经网络算法检测封面上的文字;最后,通过CRNN神经网络算法识别检测出来的文字。该系统具有操作简单、鲁棒性强的优点,可用于图书封面文字信息的自动快速采集。

Detection and recognition of book cover text helps quickly get information about the book.However,in reality,this problem is susceptible to interference from complex backgrounds,lighting,geometric distortions,and different fonts,sizes,weights,and so on.Therefore,this paper proposes a three-step algorithm to solve this problem.First,the image containing the book cover is rotated and corrected by the Hough transform;Then,the text on the cover is detected by the CRAFT neural network algorithm;Finally,the detected text is recognized by the CRNN neural network algorithm.The system has the advantages of simple operation and strong robustness,and can be used for automatic and rapid collection of text information on the cover of books.