Robust image hashing using locally linear embedding

A robust image hashing using locally linear embedding (LLE) is presented. The image is first rescaled to a fixed size. Then the image is divided into key-dependent overlapping blocks. Every blocks are reshaped to vectors and combined to the matrix. The LLE algorithm is applied to the matrix. The reconstruction weights coefficients reflect the intrinsic geometric properties of image and are invariant basically to ordinary image processing and several geometric transformation.The quantization scheme is devised to compress the coefficients matrix and form the binary image hash. Test results indicate that the proposed method is robust against content-preserving modifications such as JPEG compression, filtering and affine transformation. Hashes of different images have very low collision probability.