层次聚类结合空间金字塔的图像分类

在图像分类中，视觉词典的质量直接影响着图像分类的结果，随着用户的要求提高，K-means聚类算法所构建的视觉词典已无法满足用户对图像分类的需求，为了得到高效的视觉词汇码本，针对构建视觉词典的算法进行研究，通过K-means算法和层次聚类算法的结合来达到这一目的。混合聚类算法采用K-means算法对数据样本进行初步聚类，得到一个粗略的划分；引入信息熵的属性加权，利用信息熵度量某个属性的关键性，信息熵越大的属性对聚类结果的影响越小，计算加权后的类间欧式距离，将距离相近的两个类进行合并；在空间金字塔模型框架中，将改进的混合聚类方法应用到视觉词典的构建中。实验结果表明，结合信息熵的层次聚类算法能有效提高空间金字塔模型的分类准确率。

Image Classification of Hierarchical Clustering Combined with Spatial Pyramid

In the image classification, the quality of the visual dictionary directly affects the result of image classification. With the improvement of the user''s requirements, the visual dictionary constructed by the K-means clustering algorithm can not meet the user''s requirement for image classification. In order to obtain efficient visual Vocabulary codebook, the algorithm for constructing the visual dictionary was studied, through the K-means algorithm and hierarchical clustering algorithm to achieve this goal. Hybrid clustering algorithm used K-means algorithm to initialize the data samples; In order to use the information entropy to measure the key of an attribute, the attribute of the information entropy was weighted, if the value of information entropy was the greater, the smaller the effect was. the weighted distance between the interclass classes was calculated, and the two classes with similar distances was merged. In the framework of spatial pyramid model, the improved hybrid clustering method was applied to the construction of visual dictionary. The experimental results showed that the hierarchical clustering algorithm combined with information entropy could effectively improve the classification accuracy of spatial pyramid model.