Robust image hashing via DCT and LLE

Abstract

Image hashing is a novel technology of multimedia processing and is widely used in many applications, such as image authentication, image retrieval, image indexing, copy detection and image forensics. In this study, we propose an efficient image hashing algorithm with innovative use of discrete cosine transform (DCT) and local linear embedding (LLE). Specifically, input image is first preprocessed to build a normalized image and then color vector angles of the normalized image are extracted. Next, DCT is exploited to extract a stable feature matrix. Finally, data reduction with LLE is applied to the feature matrix and the variances of LLE results are quantized and encrypted to construct hash. Experiments with large image datasets are conducted to validate efficiency of the proposed hashing. The results show that the proposed hashing can resist normal digital operations and has good discrimination. Receiver operating characteristic (ROC) curve comparisons with state-of-the-art algorithms illustrate that the proposed hashing has better performances than the compared algorithms in classification between robustness and discrimination.