Geometric invariant features in the Radon transform domain for near-duplicate image detection

Radon transform has been widely used in content-based image representation due to its excellent geometric properties. In this paper, we propose a family of geometric invariant features based on Radon transform for near-duplicate image detection. According to the theoretical analysis between geometric operations (translation, scaling, and rotation) and Radon transform, we present a geometric invariant feature model. Based on the feature model, we developed four kinds of geometric invariant features. In addition, a uniform sampling technique is introduced to combine different features. The comprehensive performance of the combined feature is better than that of a single one. Extensive experiments show that the proposed features are robust, not only to rotation and scaling, but also to other operations, such as compression, noise contamination, blurring, illumination modification, cropping, etc., and achieve strong competitive performance compared with the state-of-the-art image features.