Perceptual Image Hashing Based on Weighted LLE and FMT

This paper presents a new image hashing technique that exploits the invariant properties of Fourier Mellin Transform (FMT) and Local Linear Embedding (LLE). First, Fourier-Mellin transform is integrated into the domain pixel blocks of the normalized image due to its invariance property in order to improve performance under geometric operations, hence generating a stable feature matrix. Then, data reduction with LLE is applied to the generated feature matrix, and the final hash is constructed after encrypting the invariances of LLE. Conducted experimental evaluation shows that our proposed hashing is robust against digital manipulations, such as rotation, scaling, cropping, JPEG compression, brightness/contrast adjustment, etc, and also reaches good discrimination. Comparisons with other state-of-the-art algorithms also reveals that our hashing outperforms its counterparts in classification performances between robustness and discriminative capability.