采用空间词袋模型的图像分类方法

Image Classification Method Based on Spatial Bag of Words Model

针对词袋模型忽略视觉单词之间的空间关联而导致分类效果降低的问题,提出一种词袋模型空间信息的构造方法,算法分别统计了图像整体与局部的直方图信息.首先将图像中具有相同视觉单词标记的图像块按照到原点的距离大小进行排序,并依次计算排序后相邻两图像块与横轴之间的夹角,进而形成各标记的全局角度直方图信息;然后计算相同标记的图像块的局部频率直方图,将二者直方图信息结合起来,完成词袋模型空间信息的构造.实验发现,采用具有空间信息的词袋模型进行图像分类,其分类结果优于其他空间算法5%以上.

To solve the problem of bag-of-words model which ignores the spatial connection between visual words, an approach for spatial information construction for bag-of-words is proposed. The histogram information in Global and local are calculated respecfly. First, the image patches with a same visual word label are sorted according to the distance to the origin. Then the angle between the line connecting two adjacent patches and the horizontal axis is computed,and the global angle histogram is established after statistical analy- sis. The global histogram is then combined with a local one which considers the local frequency histogram to construct the spatial infor- mation of bag-of-words. The experiments show that the classification result of this method is better than other algorithms more than 5%.