基于加权排序检索和视觉模式挖掘的商标识别

摘要

To recognize the logos which were contained in real-world images,this paper proposed an algorithm based on weigh-ted sorting and visual pattern mining.The algorithm firstly got preliminary recognition results using weights of feature similari-ty.Then the algorithm described a spatial relationship model between feature points and constructed stable relationships as vi-sual pattern.It applied data mining algorithm on the visual patterns to eliminate the mismatched feature pairs and obtained the final results.The paper used this algorithm and other different ones to test on FlickrLogos.Evaluation results show that visual patterns more appropriately describe logos and provide better performance than previous approaches and it achieves higher pre-cision and recall based on less sample images.%为了识别在自然条件下拍摄的图像中所包含的商标，提出了一种基于加权排序检索和视觉模式挖掘的算法。通过特征点相似度的权重大小得到商标的初步识别结果；然后建立特征点对的空间关系模型，再通过数据挖掘方法对空间位置关系所建立的视觉模式进行匹配从而删除误匹配结果，最终实现商标的识别。不同算法在数据集 FlickrLogos 上的实验结果表明，该算法利用视觉模式能更好地描述商标并且能够利用较少的样图获得较高的查准率和查全率。