基于信息熵的Logo分类问题研究

Logo classification based on information entropy

匹配是实现图像分类的关键问题,由于匹配问题的复杂性,目前还没有一个较好的解决办法。该文提出了一种基于Fourier变换和信息熵相结合的匹配算法,对Logo的分类问题进行了研究。通过Fourier变换在图像的频域中找到最佳匹配,使用相关度阈值与信息熵差比作为衡量标准。实验中选取了大量商品图像对Logo匹配问题中查准率和查全率进行了统计分析。实验结果表明,当选取适当相关度阈值与信息熵差比的情况下,该算法能有效提高商品图像按Logo的分类效果。

Image matching is a key issue in classification algorithms, but due to the complexity of the image matching, no good algorithms have been developed. This paper presents an algorithm that combines Fourier transforms and information entropy for Logo classification. The correlation ratio threshold and the entropy difference ratio threshold are used to evaluate the matching results. The Logo matching accuracy and recall ratios with the algorithm were statistically analyzed for a large number of E-goods images. The results show that the algorithm effectively classifies the Logos given the proper correlation threshold and entropy difference ratio threshold.