基于SVM的文字商标检测

Text Trademark Detection Based on the SVM

随着电子商务的兴起,越来越多的不法厂商和店家常常会有商标侵权事件,而商标中文字型商标占了很大的比重,从图片中查找出文字商标可以借鉴一些文字检测的方法。但是相对于后期加在图片上的文字来说,检测文字商标更为复杂,它是处于各种不同场景图片中,常常出现在复杂的背景上,其大小、方向、颜色都是千变万化的。本文提出了一种新的文字商标检测的算法,本算法结合了基于边缘检测的粗定位模版和SVM分类器的优势,相比于传统方法,本文提出的方法提高了文字商标检测的准确率,同时由于使用了二级过滤方法,所以算法的时间复杂度比单纯利用机器学习的方法有了很大提高。

With the development of e-commerce, more and more illegal manufactures and shopkeepers often have trademark infringement events, and a large part of trademarks are text trademarks. So finding out the text trademarks from images can be reference for some methods of text detection. But in contrast to the words added to the picture later, the detection of text trademarks are more complicated. Because they are in all kinds of scene pictures and the background are more complicated. Furthermore, their size, direction and colors are ever-changing. This paper proposes a new algorithm of text trademark detection, which combines the rough text location based on multi- scale edge detection and the advantages of SVM classifier. Compared with the traditional methods, the proposed algorithm can improves the accuracy of text trademark detection. Because of using secondary filter method, the time complexity has greatly improved than using machine learning only.