Image Hashing Algorithms Based On DWT Feature Points And Direction Histogram

Image hashing is a novel technology of multimedia security. It derives a compact representation named image hash according to visual content of input image and is widely used in many applications, such as image authentication, image forensics, copy detection and image retrieval. In general, image hashing should satisfy two basic properties:perceptual robustness and discrimination. Perceptual robustness means that image hashing should be robust to content-preserving manipulations. In other words, hashes of visually similar images should be the same or very similar. Discrimination implies that those images with different contents should have different hashes. Perceptual robustness guarantees image hashing algorithm can judge images as similar images that subjected to general digital contrast adjustment, gamma correction and other operations. Discrimination guarantees image hashing algorithm can accurately distinguish different image visual content. In addition to these two features, image hashing algorithm in a particular application must have other features. For example, the image forensics needs security, and so on.This thesis firstly analyses the effects of different color spaces on hash algorithm performance, then starting with statistical order moments and image grain direction features respectively, we introduce two technologies to extract hash. They are discrete wavelet transform (DWT) based image hashing, and direction histogram based image hashing. Specific studies are as follows.1.1 studied color space selection.Color image is a class of digital image that its number is the largest. In processing color images, most existing algorithms will generally convert a color image into YCbCr color space and take the luminance component Y for hash generation. The color space selection lacks of theoretical and experimental basis in these algorithms. Here, we discuss color space selection by evaluating classification performances of four hashing algorithms under YCbCr color space, CIE L*a*b* color space, HSV color space and HSI color space. The ROC results show that different color spaces have a great impact on the performance of the image hashing algorithm, and the regularly used YCbCr color space is not good enough in making desirable classification.2.1 proposed a DWT feature points based image hashing algorithm.After the discrete wavelet transform, the low-frequency sub-band contain most of the images information, greatly reducing the image data representing image with it. Based on the characteristics, we study the statistical order moment of DWT coefficients to construct feature points, and propose the image hashing algorithm based on DWT feature points. This algorithm converts the input image to the normalized image firstly, then converts color space, then applies DWT to gray-scale image, to construct the feature points in frequency domain. Finally, we quantize the feature points to generate the image hash. The experimental results show that the algorithm has better classification performance in both robustness and discrimination.3.1 proposed a direction histogram based image hashing algorithm.The Gabor filter is applied to the image, and its output response is maximized, when the grain direction of the image consistent with the direction of the filter. Based on this characteristic, we proposed the image hash algorithm by extracting image directions with Gabor filters. After processing the input image, we get the filter response in different directions through convolution of Gabor filters and image. Then we extract the local direction histogram of the image, and quantize the hash feature with an ordinal number of hash. Finally, we produce the decimal hash. The experimental results show that its classification performance is good. The experimental results show that the performance of receiver operating characteristic(ROC) curve are beyond the GF-LVQ hashing, CVA-DWT hashing and QSVD hashing algorithm.