A novel perception based image categorization algorithm

Image categorization plays an important role in the field of image retrieval and automatic annotation. Most existing algorithms adopted the low-level visual features to represent an image and the well-known Bag-of-words model followed by the SVM classifier was employed to fulfill the classification task. In this paper, instead of using the traditional low-level visual features, we employ the perception theory and propose a novel image categorization algorithm based on the representation of images' topological properties, i.e., an image can be represented by a low-dimensional features (e.g., Euler characteristic, the number of holes) which is called the topological properties. Given the new representations of images, a naive Bayes classifier is performed to fulfill the categorization task. The experimental results on the well-known image dataset show that based on the topological representation, the image categorization performance can be well improved.