Unsupervised image categorization with improved spectral clustering

This paper presents an approach for unsupervised image categorization which can be used in non-label image library. We extract the basic feature from image, and by using bag of words model and spatial pyramid matching we manage to improve image descriptor performance. To get better clustering performance we propose an improved spectral clustering approach and use it to achieve our image categorization object. We test our approach on two widely used image datasets. Furthermore we compare our approach to several methods and experimental results show that our approach is effective.