Reduced Complexity Image Clustering Based on Camera Fingerprints

Abstract

This work presents a reduced complexity image clustering (RCIC) algorithm that blindly groups images based on their camera fingerprint. The algorithm does not need any prior information and can be implemented without and with attraction, to refine clusters. After a camera fingerprint is estimated for each image in the data set, a fingerprint is randomly selected as reference fingerprint and a cluster is constructed using this fingerprint as centroid. The clustered fingerprints are removed from the data set and the remaining fingerprints are clustered repeating the same process. A further attraction stage can be included, in which a similar algorithm is performed using the centroids of the clusters found after the first stage. Despite its simplicity, results show that RCIC algorithm has lower computational cost than existing algorithms, while maintaining similar or even better performance. Moreover, the performance of the proposed algorithm is not affected significantly when the number of cameras in the data set is much larger than the average number of images from each camera.