基于支持向量机的线条图像语义主题自动发现方法

Automatic Semantic Topic Discovery Approach of the Line Image Based on Support Vector Machine

摘要: 提出了一种基于支持向量机分类器的线条图像语义主题自动发现方法。首先对训练图像进行分块，在对图像子块进行聚类后，得到由聚类中心构成的类集合；从每幅训练图像的注释文字中提取所有名词构成关键词集合。其次，对未标注的测试图像进行同样分块处理，计算子块与每个关键词的相关性，得到每个子块的标注词集合。最后，计算每个关键词在各个子块标注中出现的次数，取出现次数最多的关键词作为图像的语义主题。实验结果表明，所提出的方法对于线条图像的语义主题自动发现是有效的，具有比较好的性能。

Abstract: A semantic topic discovery approach of the line image, based on support vector machine, has been proposed in this paper. Firstly, the training images are divided into non-overlapping sub- blocks with same size. After clustering image sub-blocks, we obtained class set generated by cluster centers, and extracted all nouns from text annotation of each training image in order to obtain a keyword set. Secondly, the un-label testing image is also divided into non-overlapping sub-blocks as same as training images, we calculated the correlation between the sub-block and each keyword, and a keywords set for each sub-block may be obtained. Finally, the number of each keyword appearing in the each sub-block is calculated, we let the keywords with maximum to occurrences number be the semantic topics of the line image. The experimental results confirm that proposed automatic semantic topic discovery approach for line image is effective and has good performance.