Perceptual Image Hashing Based on Color Structure and Intensity Gradient

In order to improve classification performance of the hash algorithm for color images, this paper presents an hash algorithm that based on three-dimensional color structure features and luminance gradient feature. Firstly, the algorithm preprocesses the input image to produce a secondary image, and extracts the color opponent component from the secondary image. Secondly, three-dimensional external structural features are constructed by using the peak curve, mean curve, and valley curve of the color opponent component. Thirdly, the overlapping points of the peak points in different visual angles and the valley points in different visual angles are selected as the salient points. The position information of the salient points are used to construct internal features. The gradient feature is constructed by using pixel changes in four directions of Y component of luminance image in YCbCr color space. Finally, the three-dimensional color structure features and luminance gradient feature are combined and disturbed to obtain the final hash sequence. Experiment results show that the proposed scheme has preferable robustness and discrimination. Compared with some existing schemes, the proposed scheme has better classification performance, shorter hash length and less average time of hash generation, and it has good detection performance in image copy detection and image tampered detection.