Abstract: Since the cryptographic Hash function is sensitive, any slight change of the input message will influence the result of the Hash value significantly. But for the scenario of image hashing, the traditional cryptographic Hash functions will not be suitable. This is due to the fact that the images usually must undergo various manipulations and many of the manipulations are content-preserving, even though the digital representations of the images will be changed. This paper proposes a perceptual robust image hashing scheme, which can be applied in the fields such as image retrieval and authentication. Image resizing and total-variation-based filtering are firstly used to pre-process the input image for regularization. The sign relationship matrix of DCT low frequency coefficients for each image block and and corresponding neighborhood are then extracted, which can effectively reflect the distribution feature of local image content. The extracted feature matrix is finally compressed using the secret sharing mechanism to produce the final binary Hash after scrambling. The security of the scheme completely depends on the secret keys. Experiments are conducted to show the presented scheme has satisfactory robustness performance against perceptually content-preserving manipulations, e.g., JPEG compression, Gaussian low-pass filtering, and resizing, and simultaneously has very low anti-collision probability for the hashes of distinct images.