Towards efficient privacy-preserving image similarity detection

Abstract

Image similarity plays an essential role in many real world applications such as content based image retrieval (CBIR), computer vision, and near duplicate image detection. The state of the art methods are generally assumed that the content of images is not private. This reduces the utilization of these methods to work within only environments where images are publicly access. Essentially, this assumption limits more practical applications, e.g., image matching between two security agencies, where images are confidential. We address the problem of evaluating the similarity between image collections of two parties, who are reluctant to reveal their actual content. The Euclidian distance measure is used to measure distances between global hierarchal color histograms. We conduct several experiments on real image collections to demonstrate the practical value of the proposed schemes.