基于多特征和SVM的兵马俑碎片分类

Classification of Terra-Cotta Warriors fragments based on multi-feature and SVM

按照兵马俑各部位对碎片进行分类是兵马俑文物碎片拼接的重要步骤,能有效缩减自动拼接算法的搜索空间,提高拼接的准确率。由于人工的碎片分类方法工作量大,通过计算机辅助文物碎片自动分类,可以减少人工分类产生的繁重工作量。该文提出了一种基于多特征和支持向量机(SVM)的文物碎片分类方法。首先,利用尺度不变特征变换(SIFT)算法提取碎片纹理特征,在此基础上构建每幅碎片图像的词袋模型(Bo W)。其次,利用Hu不变矩提取碎片形状特征,最后,将纹理特征和形状特征结合并通过SVM进行训练,得到相应的文物碎片分类模型。实验结果表明,该方法显著提高了碎片分类的准确率。

The initial classification of fragments was an important step in the automatic splicing of Terra-Cotta Warriors fragments, which reduced the search space of the automatic splicing algorithm effectively and im- proved the accuracy of splicing. Due to the large workload of manual classification method, using the computer aided fragment classification could reduce the heavy workload of artificial classification. A new method based on multi-feature and support vector machine （SVM） for the classification of Terra-Cotta Warriors fragments is proposed. First, it used the SIFT algorithm to extract the texture features, which then was represented with histograms of features by constituting the Bag of Words model （BoW） ; Secondly, it used Hu invariant mo- ments to extract the shape features; Finally, it combined the texture features and shape features and trained SVM to get the corresponding classification model. The experimental results show that the proposed method can significantly improve the accuracy of the classification of fragments.