Image Hashing Based on CS-LBP and DCT for Copy Detection

Abstract

In order to improve the accuracy and speed of copy detection, we propose an image hashing algorithm based on local feature and global feature in this paper. Firstly, input image is converted to a normalized image. Then, wavelet decomposition is applied to the preprocessed image to produce approximate image and high-frequency information. Center symmetric local binary pattern (CS-LBP) is applied to approximate image to produce a CS-LBP texture image. The local feature is extracted by dividing blocks and selecting mean, variance, third moment, fourth moment statistics feature in CS-LBP texture image. The global feature is extracted by selecting the discrete cosine transform (DCT) coefficients of high frequency-information. Finally, local features and global features are combined to generate image hashing. Experiment with open images are carried out and the results show that the proposed algorithm is robust and discriminative. The precision rate and the recall rate (P-R) curve comparisons show that our hashing algorithm outperforms some existing hashing algorithms in copy detection.