Content-Based Image Indexing by Data Clustering and Inverse Document Frequency

In this paper we present an algorithm for creating and searching large image databases. Effective browsing and searching such collections of images based on their content is one of the most important challenges of computer science. In the presented algorithm, the process of inserting data to the database consists of several stages. In the first step interest points are generated from images by e. g. SIFT, SURF or PCA SIFT algorithms. The resulting huge number of key points is then reduced by data clustering, in our case by a novel, parameterless version of the mean shift algorithm. The reduction is achieved by subsequent operation on generated cluster centers. This algorithm has been adapted specifically for the presented method. Cluster centers are treated as terms and images as documents in the term frequency-inverse document frequency (TF-IDF) algorithm. TF-IDF algorithm allows to create an indexed image database and to fast retrieve desired images. The proposed approach is validated by numerical experiments on images with different content.